diff --git a/.idea/.gitignore b/.idea/.gitignore
new file mode 100644
index 0000000..26d3352
--- /dev/null
+++ b/.idea/.gitignore
@@ -0,0 +1,3 @@
+# Default ignored files
+/shelf/
+/workspace.xml
diff --git a/.idea/inspectionProfiles/Project_Default.xml b/.idea/inspectionProfiles/Project_Default.xml
new file mode 100644
index 0000000..6fc75ed
--- /dev/null
+++ b/.idea/inspectionProfiles/Project_Default.xml
@@ -0,0 +1,44 @@
+
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/inspectionProfiles/profiles_settings.xml b/.idea/inspectionProfiles/profiles_settings.xml
new file mode 100644
index 0000000..105ce2d
--- /dev/null
+++ b/.idea/inspectionProfiles/profiles_settings.xml
@@ -0,0 +1,6 @@
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/misc.xml b/.idea/misc.xml
new file mode 100644
index 0000000..4e1828e
--- /dev/null
+++ b/.idea/misc.xml
@@ -0,0 +1,4 @@
+
+
+
+
\ No newline at end of file
diff --git a/.idea/modules.xml b/.idea/modules.xml
new file mode 100644
index 0000000..b0315b4
--- /dev/null
+++ b/.idea/modules.xml
@@ -0,0 +1,8 @@
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/yolov5_obb.iml b/.idea/yolov5_obb.iml
new file mode 100644
index 0000000..cc2afc0
--- /dev/null
+++ b/.idea/yolov5_obb.iml
@@ -0,0 +1,14 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.vs/ProjectSettings.json b/.vs/ProjectSettings.json
new file mode 100644
index 0000000..f8b4888
--- /dev/null
+++ b/.vs/ProjectSettings.json
@@ -0,0 +1,3 @@
+{
+ "CurrentProjectSetting": null
+}
\ No newline at end of file
diff --git a/.vs/VSWorkspaceState.json b/.vs/VSWorkspaceState.json
new file mode 100644
index 0000000..5f25a06
--- /dev/null
+++ b/.vs/VSWorkspaceState.json
@@ -0,0 +1,7 @@
+{
+ "ExpandedNodes": [
+ ""
+ ],
+ "SelectedNode": "\\train.py",
+ "PreviewInSolutionExplorer": false
+}
\ No newline at end of file
diff --git a/.vs/slnx.sqlite b/.vs/slnx.sqlite
new file mode 100644
index 0000000..ef51a48
Binary files /dev/null and b/.vs/slnx.sqlite differ
diff --git a/.vs/yolov5_obb/v15/.suo b/.vs/yolov5_obb/v15/.suo
new file mode 100644
index 0000000..c3260b5
Binary files /dev/null and b/.vs/yolov5_obb/v15/.suo differ
diff --git a/.vs/yolov5_obb/v15/Browse.VC.db b/.vs/yolov5_obb/v15/Browse.VC.db
new file mode 100644
index 0000000..1e94395
Binary files /dev/null and b/.vs/yolov5_obb/v15/Browse.VC.db differ
diff --git a/Arial.ttf b/Arial.ttf
new file mode 100644
index 0000000..ab68fb1
Binary files /dev/null and b/Arial.ttf differ
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
new file mode 100644
index 0000000..fcceba2
--- /dev/null
+++ b/CONTRIBUTING.md
@@ -0,0 +1,94 @@
+## Contributing to YOLOv5 🚀
+
+We love your input! We want to make contributing to YOLOv5 as easy and transparent as possible, whether it's:
+
+- Reporting a bug
+- Discussing the current state of the code
+- Submitting a fix
+- Proposing a new feature
+- Becoming a maintainer
+
+YOLOv5 works so well due to our combined community effort, and for every small improvement you contribute you will be
+helping push the frontiers of what's possible in AI 😃!
+
+## Submitting a Pull Request (PR) 🛠️
+
+Submitting a PR is easy! This example shows how to submit a PR for updating `requirements.txt` in 4 steps:
+
+### 1. Select File to Update
+
+Select `requirements.txt` to update by clicking on it in GitHub.
+
+
+### 2. Click 'Edit this file'
+
+Button is in top-right corner.
+
+
+### 3. Make Changes
+
+Change `matplotlib` version from `3.2.2` to `3.3`.
+
+
+### 4. Preview Changes and Submit PR
+
+Click on the **Preview changes** tab to verify your updates. At the bottom of the screen select 'Create a **new branch**
+for this commit', assign your branch a descriptive name such as `fix/matplotlib_version` and click the green **Propose
+changes** button. All done, your PR is now submitted to YOLOv5 for review and approval 😃!
+
+
+### PR recommendations
+
+To allow your work to be integrated as seamlessly as possible, we advise you to:
+
+- ✅ Verify your PR is **up-to-date with upstream/master.** If your PR is behind upstream/master an
+ automatic [GitHub actions](https://github.com/ultralytics/yolov5/blob/master/.github/workflows/rebase.yml) rebase may
+ be attempted by including the /rebase command in a comment body, or by running the following code, replacing 'feature'
+ with the name of your local branch:
+
+ ```bash
+ git remote add upstream https://github.com/ultralytics/yolov5.git
+ git fetch upstream
+ git checkout feature # <----- replace 'feature' with local branch name
+ git merge upstream/master
+ git push -u origin -f
+ ```
+
+- ✅ Verify all Continuous Integration (CI) **checks are passing**.
+- ✅ Reduce changes to the absolute **minimum** required for your bug fix or feature addition. _"It is not daily increase
+ but daily decrease, hack away the unessential. The closer to the source, the less wastage there is."_ — Bruce Lee
+
+## Submitting a Bug Report 🐛
+
+If you spot a problem with YOLOv5 please submit a Bug Report!
+
+For us to start investigating a possible problem we need to be able to reproduce it ourselves first. We've created a few
+short guidelines below to help users provide what we need in order to get started.
+
+When asking a question, people will be better able to provide help if you provide **code** that they can easily
+understand and use to **reproduce** the problem. This is referred to by community members as creating
+a [minimum reproducible example](https://stackoverflow.com/help/minimal-reproducible-example). Your code that reproduces
+the problem should be:
+
+* ✅ **Minimal** – Use as little code as possible that still produces the same problem
+* ✅ **Complete** – Provide **all** parts someone else needs to reproduce your problem in the question itself
+* ✅ **Reproducible** – Test the code you're about to provide to make sure it reproduces the problem
+
+In addition to the above requirements, for [Ultralytics](https://ultralytics.com/) to provide assistance your code
+should be:
+
+* ✅ **Current** – Verify that your code is up-to-date with current
+ GitHub [master](https://github.com/ultralytics/yolov5/tree/master), and if necessary `git pull` or `git clone` a new
+ copy to ensure your problem has not already been resolved by previous commits.
+* ✅ **Unmodified** – Your problem must be reproducible without any modifications to the codebase in this
+ repository. [Ultralytics](https://ultralytics.com/) does not provide support for custom code ⚠️.
+
+If you believe your problem meets all of the above criteria, please close this issue and raise a new one using the 🐛 **
+Bug Report** [template](https://github.com/ultralytics/yolov5/issues/new/choose) and providing
+a [minimum reproducible example](https://stackoverflow.com/help/minimal-reproducible-example) to help us better
+understand and diagnose your problem.
+
+## License
+
+By contributing, you agree that your contributions will be licensed under
+the [GPL-3.0 license](https://choosealicense.com/licenses/gpl-3.0/)
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..9a55005
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,64 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+
+# Start FROM Nvidia PyTorch image https://ngc.nvidia.com/catalog/containers/nvidia:pytorch
+FROM nvcr.io/nvidia/pytorch:21.10-py3
+
+# Install linux packages
+RUN apt update && apt install -y zip htop screen libgl1-mesa-glx
+
+# Install python dependencies
+COPY requirements.txt .
+RUN python -m pip install --upgrade pip
+RUN pip uninstall -y nvidia-tensorboard nvidia-tensorboard-plugin-dlprof
+RUN pip install --no-cache -r requirements.txt coremltools onnx gsutil notebook wandb>=0.12.2
+RUN pip install --no-cache -U torch torchvision numpy Pillow
+# RUN pip install --no-cache torch==1.10.0+cu113 torchvision==0.11.1+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
+
+# Create working directory
+RUN mkdir -p /usr/src/app
+WORKDIR /usr/src/app
+
+# Copy contents
+COPY . /usr/src/app
+
+# Downloads to user config dir
+ADD https://ultralytics.com/assets/Arial.ttf /root/.config/Ultralytics/
+
+# Set environment variables
+# ENV HOME=/usr/src/app
+
+
+# Usage Examples -------------------------------------------------------------------------------------------------------
+
+# Build and Push
+# t=ultralytics/yolov5:latest && sudo docker build -t $t . && sudo docker push $t
+
+# Pull and Run
+# t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host --gpus all $t
+
+# Pull and Run with local directory access
+# t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host --gpus all -v "$(pwd)"/datasets:/usr/src/datasets $t
+
+# Kill all
+# sudo docker kill $(sudo docker ps -q)
+
+# Kill all image-based
+# sudo docker kill $(sudo docker ps -qa --filter ancestor=ultralytics/yolov5:latest)
+
+# Bash into running container
+# sudo docker exec -it 5a9b5863d93d bash
+
+# Bash into stopped container
+# id=$(sudo docker ps -qa) && sudo docker start $id && sudo docker exec -it $id bash
+
+# Clean up
+# docker system prune -a --volumes
+
+# Update Ubuntu drivers
+# https://www.maketecheasier.com/install-nvidia-drivers-ubuntu/
+
+# DDP test
+# python -m torch.distributed.run --nproc_per_node 2 --master_port 1 train.py --epochs 3
+
+# GCP VM from Image
+# docker.io/ultralytics/yolov5:latest
diff --git a/LICENSE b/LICENSE
new file mode 100644
index 0000000..92b370f
--- /dev/null
+++ b/LICENSE
@@ -0,0 +1,674 @@
+GNU GENERAL PUBLIC LICENSE
+ Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc.
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+ Preamble
+
+ The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+ The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works. By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users. We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors. You can apply it to
+your programs, too.
+
+ When we speak of free software, we are referring to freedom, not
+price. Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+ To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights. Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+ For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received. You must make sure that they, too, receive
+or can get the source code. And you must show them these terms so they
+know their rights.
+
+ Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+ For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software. For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+ Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so. This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software. The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable. Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products. If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+ Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary. To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+ The precise terms and conditions for copying, distribution and
+modification follow.
+
+ TERMS AND CONDITIONS
+
+ 0. Definitions.
+
+ "This License" refers to version 3 of the GNU General Public License.
+
+ "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+ "The Program" refers to any copyrightable work licensed under this
+License. Each licensee is addressed as "you". "Licensees" and
+"recipients" may be individuals or organizations.
+
+ To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy. The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+ A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+ To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy. Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+ To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies. Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+ An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License. If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+ 1. Source Code.
+
+ The "source code" for a work means the preferred form of the work
+for making modifications to it. "Object code" means any non-source
+form of a work.
+
+ A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+ The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form. A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+ The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities. However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work. For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+ The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+ The Corresponding Source for a work in source code form is that
+same work.
+
+ 2. Basic Permissions.
+
+ All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met. This License explicitly affirms your unlimited
+permission to run the unmodified Program. The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work. This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+ You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force. You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright. Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+ Conveying under any other circumstances is permitted solely under
+the conditions stated below. Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+ 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+ No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+ When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+ 4. Conveying Verbatim Copies.
+
+ You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+ You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+ 5. Conveying Modified Source Versions.
+
+ You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+ a) The work must carry prominent notices stating that you modified
+ it, and giving a relevant date.
+
+ b) The work must carry prominent notices stating that it is
+ released under this License and any conditions added under section
+ 7. This requirement modifies the requirement in section 4 to
+ "keep intact all notices".
+
+ c) You must license the entire work, as a whole, under this
+ License to anyone who comes into possession of a copy. This
+ License will therefore apply, along with any applicable section 7
+ additional terms, to the whole of the work, and all its parts,
+ regardless of how they are packaged. This License gives no
+ permission to license the work in any other way, but it does not
+ invalidate such permission if you have separately received it.
+
+ d) If the work has interactive user interfaces, each must display
+ Appropriate Legal Notices; however, if the Program has interactive
+ interfaces that do not display Appropriate Legal Notices, your
+ work need not make them do so.
+
+ A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit. Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+ 6. Conveying Non-Source Forms.
+
+ You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+ a) Convey the object code in, or embodied in, a physical product
+ (including a physical distribution medium), accompanied by the
+ Corresponding Source fixed on a durable physical medium
+ customarily used for software interchange.
+
+ b) Convey the object code in, or embodied in, a physical product
+ (including a physical distribution medium), accompanied by a
+ written offer, valid for at least three years and valid for as
+ long as you offer spare parts or customer support for that product
+ model, to give anyone who possesses the object code either (1) a
+ copy of the Corresponding Source for all the software in the
+ product that is covered by this License, on a durable physical
+ medium customarily used for software interchange, for a price no
+ more than your reasonable cost of physically performing this
+ conveying of source, or (2) access to copy the
+ Corresponding Source from a network server at no charge.
+
+ c) Convey individual copies of the object code with a copy of the
+ written offer to provide the Corresponding Source. This
+ alternative is allowed only occasionally and noncommercially, and
+ only if you received the object code with such an offer, in accord
+ with subsection 6b.
+
+ d) Convey the object code by offering access from a designated
+ place (gratis or for a charge), and offer equivalent access to the
+ Corresponding Source in the same way through the same place at no
+ further charge. You need not require recipients to copy the
+ Corresponding Source along with the object code. If the place to
+ copy the object code is a network server, the Corresponding Source
+ may be on a different server (operated by you or a third party)
+ that supports equivalent copying facilities, provided you maintain
+ clear directions next to the object code saying where to find the
+ Corresponding Source. Regardless of what server hosts the
+ Corresponding Source, you remain obligated to ensure that it is
+ available for as long as needed to satisfy these requirements.
+
+ e) Convey the object code using peer-to-peer transmission, provided
+ you inform other peers where the object code and Corresponding
+ Source of the work are being offered to the general public at no
+ charge under subsection 6d.
+
+ A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+ A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling. In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage. For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product. A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+ "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source. The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+ If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information. But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+ The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed. Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+ Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+ 7. Additional Terms.
+
+ "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law. If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+ When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it. (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.) You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+ Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+ a) Disclaiming warranty or limiting liability differently from the
+ terms of sections 15 and 16 of this License; or
+
+ b) Requiring preservation of specified reasonable legal notices or
+ author attributions in that material or in the Appropriate Legal
+ Notices displayed by works containing it; or
+
+ c) Prohibiting misrepresentation of the origin of that material, or
+ requiring that modified versions of such material be marked in
+ reasonable ways as different from the original version; or
+
+ d) Limiting the use for publicity purposes of names of licensors or
+ authors of the material; or
+
+ e) Declining to grant rights under trademark law for use of some
+ trade names, trademarks, or service marks; or
+
+ f) Requiring indemnification of licensors and authors of that
+ material by anyone who conveys the material (or modified versions of
+ it) with contractual assumptions of liability to the recipient, for
+ any liability that these contractual assumptions directly impose on
+ those licensors and authors.
+
+ All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10. If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term. If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+ If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+ Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+ 8. Termination.
+
+ You may not propagate or modify a covered work except as expressly
+provided under this License. Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+ However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+ Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+ Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License. If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+ 9. Acceptance Not Required for Having Copies.
+
+ You are not required to accept this License in order to receive or
+run a copy of the Program. Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance. However,
+nothing other than this License grants you permission to propagate or
+modify any covered work. These actions infringe copyright if you do
+not accept this License. Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+ 10. Automatic Licensing of Downstream Recipients.
+
+ Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License. You are not responsible
+for enforcing compliance by third parties with this License.
+
+ An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations. If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+ You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License. For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+ 11. Patents.
+
+ A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based. The
+work thus licensed is called the contributor's "contributor version".
+
+ A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version. For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+ Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+ In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement). To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+ If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients. "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+ If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+ A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License. You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+ Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+ 12. No Surrender of Others' Freedom.
+
+ If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License. If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all. For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+ 13. Use with the GNU Affero General Public License.
+
+ Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work. The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+ 14. Revised Versions of this License.
+
+ The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time. Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+ Each version is given a distinguishing version number. If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation. If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+ If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+ Later license versions may give you additional or different
+permissions. However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+ 15. Disclaimer of Warranty.
+
+ THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+ 16. Limitation of Liability.
+
+ IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+ 17. Interpretation of Sections 15 and 16.
+
+ If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+ END OF TERMS AND CONDITIONS
+
+ How to Apply These Terms to Your New Programs
+
+ If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+ To do so, attach the following notices to the program. It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+
+ Copyright (C)
+
+ This program is free software: you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation, either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see .
+
+Also add information on how to contact you by electronic and paper mail.
+
+ If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+ Copyright (C)
+ This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+ This is free software, and you are welcome to redistribute it
+ under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License. Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+ You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+.
+
+ The GNU General Public License does not permit incorporating your program
+into proprietary programs. If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library. If this is what you want to do, use the GNU Lesser General
+Public License instead of this License. But first, please read
+.
diff --git a/__pycache__/val.cpython-37.pyc b/__pycache__/val.cpython-37.pyc
new file mode 100644
index 0000000..d91414d
Binary files /dev/null and b/__pycache__/val.cpython-37.pyc differ
diff --git a/dataset/dataset_demo/images/P0032.png b/dataset/dataset_demo/images/P0032.png
new file mode 100644
index 0000000..923ade1
Binary files /dev/null and b/dataset/dataset_demo/images/P0032.png differ
diff --git a/dataset/dataset_demo/imgnamefile.txt b/dataset/dataset_demo/imgnamefile.txt
new file mode 100644
index 0000000..b5de060
--- /dev/null
+++ b/dataset/dataset_demo/imgnamefile.txt
@@ -0,0 +1 @@
+P0032
diff --git a/dataset/dataset_demo/labelTxt/P0032.txt b/dataset/dataset_demo/labelTxt/P0032.txt
new file mode 100644
index 0000000..c7fbae5
--- /dev/null
+++ b/dataset/dataset_demo/labelTxt/P0032.txt
@@ -0,0 +1,56 @@
+1686.0 1517.0 1695.0 1511.0 1711.0 1535.0 1700.0 1541.0 large-vehicle 1
+1753.0 428.0 1745.0 417.0 1785.0 391.0 1792.0 402.0 large-vehicle 0
+339.0 85.0 334.0 89.0 326.0 79.0 331.0 74.0 small-vehicle 1
+347.0 77.0 340.0 81.0 330.0 67.0 337.0 62.0 small-vehicle 1
+366.0 68.0 358.0 72.0 347.0 56.0 355.0 50.0 small-vehicle 1
+386.0 56.0 376.0 63.0 365.0 47.0 374.0 41.0 small-vehicle 1
+223.0 122.0 216.0 113.0 228.0 106.0 234.0 114.0 small-vehicle 1
+167.0 102.0 158.0 107.0 147.0 89.0 156.0 82.0 small-vehicle 1
+218.0 89.0 224.0 84.0 236.0 101.0 227.0 106.0 small-vehicle 1
+280.0 131.0 272.0 136.0 264.0 121.0 270.0 116.0 small-vehicle 1
+311.0 128.0 303.0 132.0 292.0 118.0 299.0 114.0 small-vehicle 1
+266.0 81.0 258.0 85.0 247.0 69.0 254.0 62.0 small-vehicle 1
+276.0 76.0 267.0 81.0 260.0 65.0 268.0 60.0 small-vehicle 1
+285.0 70.0 276.0 74.0 268.0 58.0 276.0 54.0 small-vehicle 1
+300.0 59.0 292.0 64.0 283.0 49.0 290.0 44.0 small-vehicle 1
+309.0 55.0 301.0 59.0 294.0 44.0 301.0 40.0 small-vehicle 1
+326.0 42.0 319.0 47.0 308.0 31.0 316.0 26.0 small-vehicle 1
+1054.0 977.0 1166.0 914.0 1242.0 1034.0 1134.0 1096.0 plane 0
+1299.0 800.0 1412.0 740.0 1497.0 887.0 1382.0 948.0 plane 0
+1555.0 680.0 1659.0 610.0 1745.0 738.0 1632.0 801.0 plane 0
+1794.0 532.0 1907.0 468.0 1981.0 592.0 1868.0 659.0 plane 0
+2211.0 319.0 2316.0 252.0 2395.0 374.0 2298.0 442.0 plane 0
+2042.0 473.0 2151.0 398.0 2238.0 525.0 2133.0 592.0 plane 0
+126.0 1450.0 203.0 1562.0 74.0 1648.0 7.0 1536.0 plane 0
+850.0 1154.0 954.0 1087.0 1027.0 1205.0 922.0 1268.0 plane 0
+692.0 1249.0 796.0 1181.0 868.0 1299.0 762.0 1364.0 plane 0
+288.0 1495.0 395.0 1432.0 468.0 1555.0 359.0 1616.0 plane 0
+259.0 192.0 254.0 183.0 270.0 174.0 274.0 182.0 small-vehicle 1
+268.0 206.0 264.0 199.0 281.0 187.0 286.0 196.0 small-vehicle 1
+279.0 225.0 272.0 216.0 288.0 206.0 294.0 213.0 small-vehicle 1
+627.0 857.0 639.0 850.0 665.0 891.0 654.0 899.0 large-vehicle 0
+490.0 516.0 495.0 525.0 475.0 537.0 469.0 527.0 small-vehicle 1
+605.0 740.0 599.0 732.0 613.0 719.0 620.0 727.0 small-vehicle 1
+633.0 688.0 622.0 693.0 611.0 672.0 621.0 666.0 large-vehicle 1
+483.0 486.0 488.0 496.0 457.0 514.0 451.0 504.0 large-vehicle 1
+553.0 651.0 546.0 643.0 559.0 632.0 565.0 639.0 small-vehicle 1
+555.0 666.0 547.0 658.0 565.0 646.0 570.0 655.0 small-vehicle 1
+161.0 1416.0 171.0 1408.0 201.0 1452.0 190.0 1458.0 large-vehicle 0
+200.0 1461.0 212.0 1455.0 240.0 1500.0 230.0 1507.0 large-vehicle 0
+372.0 1357.0 364.0 1341.0 411.0 1316.0 418.0 1331.0 large-vehicle 0
+336.0 1250.0 333.0 1264.0 284.0 1252.0 285.0 1237.0 large-vehicle 0
+324.0 847.0 333.0 860.0 286.0 887.0 278.0 876.0 large-vehicle 0
+318.0 797.0 329.0 789.0 358.0 829.0 346.0 837.0 large-vehicle 0
+305.0 231.0 309.0 240.0 293.0 247.0 288.0 240.0 small-vehicle 1
+431.0 31.0 426.0 35.0 414.0 20.0 420.0 13.0 small-vehicle 1
+395.0 29.0 401.0 22.0 411.0 40.0 405.0 44.0 small-vehicle 1
+412.0 338.0 404.0 339.0 401.0 319.0 407.0 318.0 small-vehicle 1
+367.0 341.0 371.0 349.0 360.0 357.0 355.0 349.0 small-vehicle 1
+362.0 333.0 366.0 341.0 352.0 349.0 347.0 343.0 small-vehicle 1
+346.0 331.0 339.0 324.0 350.0 315.0 356.0 323.0 small-vehicle 1
+336.0 318.0 329.0 311.0 342.0 299.0 348.0 308.0 small-vehicle 1
+324.0 299.0 318.0 292.0 330.0 284.0 334.0 292.0 small-vehicle 1
+309.0 274.0 305.0 266.0 316.0 258.0 323.0 267.0 small-vehicle 1
+311.0 249.0 316.0 258.0 303.0 264.0 298.0 258.0 small-vehicle 1
+297.0 223.0 303.0 230.0 288.0 239.0 284.0 232.0 small-vehicle 1
+550.0 1332.0 668.0 1263.0 742.0 1386.0 623.0 1457.0 plane 0
diff --git a/detect.py b/detect.py
new file mode 100644
index 0000000..3ba89b2
--- /dev/null
+++ b/detect.py
@@ -0,0 +1,251 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Run inference on images, videos, directories, streams, etc.
+
+Usage:
+ $ python path/to/detect.py --weights yolov5s.pt --source 0 # webcam
+ img.jpg # image
+ vid.mp4 # video
+ path/ # directory
+ path/*.jpg # glob
+ 'https://youtu.be/Zgi9g1ksQHc' # YouTube
+ 'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP stream
+"""
+
+import argparse
+import os
+import sys
+from pathlib import Path
+
+import cv2
+import torch
+import torch.backends.cudnn as cudnn
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[0] # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+ sys.path.append(str(ROOT)) # add ROOT to PATH
+ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
+
+from models.common import DetectMultiBackend
+from utils.datasets import IMG_FORMATS, VID_FORMATS, LoadImages, LoadStreams
+from utils.general import (LOGGER, check_file, check_img_size, check_imshow, check_requirements, colorstr,
+ increment_path, non_max_suppression, non_max_suppression_obb, print_args, scale_coords, scale_polys, strip_optimizer, xyxy2xywh)
+from utils.plots import Annotator, colors, save_one_box
+from utils.torch_utils import select_device, time_sync
+from utils.rboxs_utils import poly2rbox, rbox2poly
+
+
+@torch.no_grad()
+def run(weights=ROOT / 'yolov5s.pt', # model.pt path(s)
+ source=ROOT / 'data/images', # file/dir/URL/glob, 0 for webcam
+ imgsz=(640, 640), # inference size (height, width)
+ conf_thres=0.25, # confidence threshold
+ iou_thres=0.45, # NMS IOU threshold
+ max_det=1000, # maximum detections per image
+ device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
+ view_img=False, # show results
+ save_txt=False, # save results to *.txt
+ save_conf=False, # save confidences in --save-txt labels
+ save_crop=False, # save cropped prediction boxes
+ nosave=False, # do not save images/videos
+ classes=None, # filter by class: --class 0, or --class 0 2 3
+ agnostic_nms=False, # class-agnostic NMS
+ augment=False, # augmented inference
+ visualize=False, # visualize features
+ update=False, # update all models
+ project=ROOT / 'runs/detect', # save results to project/name
+ name='exp', # save results to project/name
+ exist_ok=False, # existing project/name ok, do not increment
+ line_thickness=3, # bounding box thickness (pixels)
+ hide_labels=False, # hide labels
+ hide_conf=False, # hide confidences
+ half=False, # use FP16 half-precision inference
+ dnn=False, # use OpenCV DNN for ONNX inference
+ ):
+ source = str(source)
+ save_img = not nosave and not source.endswith('.txt') # save inference images
+ is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
+ is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))
+ webcam = source.isnumeric() or source.endswith('.txt') or (is_url and not is_file)
+ if is_url and is_file:
+ source = check_file(source) # download
+
+ # Directories
+ save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
+ (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
+
+ # Load model
+ device = select_device(device)
+ model = DetectMultiBackend(weights, device=device, dnn=dnn)
+ stride, names, pt, jit, onnx, engine = model.stride, model.names, model.pt, model.jit, model.onnx, model.engine
+ imgsz = check_img_size(imgsz, s=stride) # check image size
+
+ # Half
+ half &= (pt or jit or engine) and device.type != 'cpu' # half precision only supported by PyTorch on CUDA
+ if pt or jit:
+ model.model.half() if half else model.model.float()
+
+ # Dataloader
+ if webcam:
+ view_img = check_imshow()
+ cudnn.benchmark = True # set True to speed up constant image size inference
+ dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt)
+ bs = len(dataset) # batch_size
+ else:
+ dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt)
+ bs = 1 # batch_size
+ vid_path, vid_writer = [None] * bs, [None] * bs
+
+ # Run inference
+ model.warmup(imgsz=(1, 3, *imgsz), half=half) # warmup
+ dt, seen = [0.0, 0.0, 0.0], 0
+ for path, im, im0s, vid_cap, s in dataset:
+ t1 = time_sync()
+ im = torch.from_numpy(im).to(device)
+ im = im.half() if half else im.float() # uint8 to fp16/32
+ im /= 255 # 0 - 255 to 0.0 - 1.0
+ if len(im.shape) == 3:
+ im = im[None] # expand for batch dim
+ t2 = time_sync()
+ dt[0] += t2 - t1
+
+ # Inference
+ visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False
+ pred = model(im, augment=augment, visualize=visualize)
+ t3 = time_sync()
+ dt[1] += t3 - t2
+
+ # NMS
+ # pred: list*(n, [xylsθ, conf, cls]) θ ∈ [-pi/2, pi/2)
+ pred = non_max_suppression_obb(pred, conf_thres, iou_thres, classes, agnostic_nms, multi_label=True, max_det=max_det)
+ dt[2] += time_sync() - t3
+
+ # Second-stage classifier (optional)
+ # pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)
+
+ # Process predictions
+ for i, det in enumerate(pred): # per image
+ pred_poly = rbox2poly(det[:, :5]) # (n, [x1 y1 x2 y2 x3 y3 x4 y4])
+ seen += 1
+ if webcam: # batch_size >= 1
+ p, im0, frame = path[i], im0s[i].copy(), dataset.count
+ s += f'{i}: '
+ else:
+ p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)
+
+ p = Path(p) # to Path
+ save_path = str(save_dir / p.name) # im.jpg
+ txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # im.txt
+ s += '%gx%g ' % im.shape[2:] # print string
+ gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
+ imc = im0.copy() if save_crop else im0 # for save_crop
+ annotator = Annotator(im0, line_width=line_thickness, example=str(names))
+ if len(det):
+ # Rescale polys from img_size to im0 size
+ # det[:, :4] = scale_coords(im.shape[2:], det[:, :4], im0.shape).round()
+ pred_poly = scale_polys(im.shape[2:], pred_poly, im0.shape)
+ det = torch.cat((pred_poly, det[:, -2:]), dim=1) # (n, [poly conf cls])
+
+ # Print results
+ for c in det[:, -1].unique():
+ n = (det[:, -1] == c).sum() # detections per class
+ s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
+
+ # Write results
+ for *poly, conf, cls in reversed(det):
+ if save_txt: # Write to file
+ # xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
+ # poly = poly.tolist()
+ line = (cls, *poly, conf) if save_conf else (cls, *poly) # label format
+ with open(txt_path + '.txt', 'a') as f:
+ f.write(('%g ' * len(line)).rstrip() % line + '\n')
+
+ if save_img or save_crop or view_img: # Add poly to image
+ c = int(cls) # integer class
+ label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}')
+ # annotator.box_label(xyxy, label, color=colors(c, True))
+ annotator.poly_label(poly, label, color=colors(c, True))
+ if save_crop: # Yolov5-obb doesn't support it yet
+ # save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)
+ pass
+
+ # Print time (inference-only)
+ LOGGER.info(f'{s}Done. ({t3 - t2:.3f}s)')
+
+ # Stream results
+ im0 = annotator.result()
+ if view_img:
+ cv2.imshow(str(p), im0)
+ cv2.waitKey(1) # 1 millisecond
+
+ # Save results (image with detections)
+ if save_img:
+ if dataset.mode == 'image':
+ cv2.imwrite(save_path, im0)
+ else: # 'video' or 'stream'
+ if vid_path[i] != save_path: # new video
+ vid_path[i] = save_path
+ if isinstance(vid_writer[i], cv2.VideoWriter):
+ vid_writer[i].release() # release previous video writer
+ if vid_cap: # video
+ fps = vid_cap.get(cv2.CAP_PROP_FPS)
+ w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+ h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
+ else: # stream
+ fps, w, h = 30, im0.shape[1], im0.shape[0]
+ save_path += '.mp4'
+ vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
+ vid_writer[i].write(im0)
+
+ # Print results
+ t = tuple(x / seen * 1E3 for x in dt) # speeds per image
+ LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)
+ if save_txt or save_img:
+ s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
+ LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
+ if update:
+ strip_optimizer(weights) # update model (to fix SourceChangeWarning)
+
+
+def parse_opt():
+ parser = argparse.ArgumentParser()
+ parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'data/HRSC2016/best.pt', help='model path(s)')
+ parser.add_argument('--source', type=str, default='data/HRSC2016/test1/images', help='file/dir/URL/glob, 0 for webcam')
+ parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[768], help='inference size h,w')
+ parser.add_argument('--conf-thres', type=float, default=0.1, help='confidence threshold')
+ parser.add_argument('--iou-thres', type=float, default=0.4, help='NMS IoU threshold')
+ parser.add_argument('--max-det', type=int, default=1000, help='maximum detections per image')
+ parser.add_argument('--device', default='0', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+ parser.add_argument('--view-img', action='store_true', help='show results')
+ parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
+ parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
+ parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
+ parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
+ parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --classes 0, or --classes 0 2 3')
+ parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
+ parser.add_argument('--augment', action='store_true', help='augmented inference')
+ parser.add_argument('--visualize', action='store_true', help='visualize features')
+ parser.add_argument('--update', action='store_true', help='update all models')
+ parser.add_argument('--project', default='runs/detect', help='save results to project/name')
+ parser.add_argument('--name', default='exp', help='save results to project/name')
+ parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
+ parser.add_argument('--line-thickness', default=2, type=int, help='bounding box thickness (pixels)')
+ parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
+ parser.add_argument('--hide-conf', default=True, action='store_true', help='hide confidences')
+ parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
+ parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
+ opt = parser.parse_args()
+ opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand
+ print_args(FILE.stem, opt)
+ return opt
+
+
+def main(opt):
+ check_requirements(exclude=('tensorboard', 'thop'))
+ run(**vars(opt))
+
+
+if __name__ == "__main__":
+ opt = parse_opt()
+ main(opt)
diff --git a/docs/ChangeLog.md b/docs/ChangeLog.md
new file mode 100644
index 0000000..e6acb81
--- /dev/null
+++ b/docs/ChangeLog.md
@@ -0,0 +1,21 @@
+
+
+# Updates
+**[2022/1/7]**
+1. Update yolov5 base version to [Releases v6.0](https://github.com/ultralytics/yolov5/releases/tag/v6.0).
+2. Rebuild the obb-label pre/post-process code. That means **Faster and Stronger** in training/validation/testing.
+
+Model| Training Dataset | BatchSize | epochs |GPU | Time Cost |OBB mAPtest @0.5| fps |
+---- | ----- | ------ | ----- | ----- | ----- | ----- | ----- |
+yolov5m-old | DOTAv1.5train_subsize1024_gap200_rate1.0|75 |300 |2080Ti |96h |68.36 |20 |
+**yolov5m-new**| DOTAv1.5train_subsize1024_gap200_rate1.0|75 |300 |2080Ti |**15h**|**73.19**|**59** |
+
+3. Some Bugs Fixed.
+
+|Bug | Fixed | Describe
+|---- |------ | ------
+|Don't support validation | ✔ | Support hbb validation in training, which is faster than obb validation|
+|Don't support single class training | ✔ | But it will get weaker results than **'nc=2'** |
+|Image must meets Height=Width | ✔ | - |
+
+4. support obb_nms gpu version.
diff --git a/docs/GetStart.md b/docs/GetStart.md
new file mode 100644
index 0000000..6418ed4
--- /dev/null
+++ b/docs/GetStart.md
@@ -0,0 +1,190 @@
+# Getting Started
+
+This page provides basic usage about yolov5-obb. For installation instructions, please see [install.md](./install.md).
+
+# Train a model
+
+**1. Prepare custom dataset files**
+
+1.1 Make sure the labels format is [poly classname diffcult], e.g., You can set **diffcult=0**
+```
+ x1 y1 x2 y2 x3 y3 x4 y4 classname diffcult
+
+1686.0 1517.0 1695.0 1511.0 1711.0 1535.0 1700.0 1541.0 large-vehicle 1
+```
+![image](https://user-images.githubusercontent.com/72599120/159213229-b7c2fc5c-b140-4f10-9af8-2cbc405b0cd3.png)
+
+
+1.2 Split the dataset.
+```shell
+cd yolov5_obb
+python DOTA_devkit/ImgSplit_multi_process.py
+```
+or Use the orignal dataset.
+```shell
+cd yolov5_obb
+```
+
+1.3 Make sure your dataset structure same as:
+```
+parent
+├── yolov5
+└── datasets
+ └── DOTAv1.5
+ ├── train_split_rate1.0_subsize1024_gap200
+ ├── train_split_rate1.0_subsize1024_gap200
+ └── test_split_rate1.0_subsize1024_gap200
+ ├── images
+ |────1.jpg
+ |────...
+ └────10000.jpg
+ ├── labelTxt
+ |────1.txt
+ |────...
+ └────10000.txt
+
+```
+
+**Note:**
+* DOTA is a high resolution image dataset, so it needs to be splited before training/testing to get better performance.
+
+**2. Train**
+
+2.1 Train with specified GPUs. (for example with GPU=3)
+
+```shell
+python train.py --device 3
+```
+
+2.2 Train with multiple(4) GPUs. (DDP Mode)
+
+```shell
+python -m torch.distributed.launch --nproc_per_node 4 train.py --device 0,1,2,3
+```
+
+2.3 Train the orignal dataset demo.
+```shell
+python train.py \
+ --weights 'weights/yolov5n_s_m_l_x.pt' \
+ --data 'data/yolov5obb_demo.yaml' \
+ --hyp 'data/hyps/obb/hyp.finetune_dota.yaml' \
+ --epochs 10 \
+ --batch-size 1 \
+ --img 1024 \
+ --device 0
+```
+
+2.4 Train the splited dataset demo.
+```shell
+python train.py \
+ --weights 'weights/yolov5n_s_m_l_x.pt' \
+ --data 'data/yolov5obb_demo_split.yaml' \
+ --hyp 'data/hyps/obb/hyp.finetune_dota.yaml' \
+ --epochs 10 \
+ --batch-size 2 \
+ --img 1024 \
+ --device 0
+```
+
+# Inferenece with pretrained models. (Splited Dataset)
+This repo provides the validation/testing scripts to evaluate the trained model.
+
+Examples:
+
+Assume that you have already downloaded the checkpoints to `runs/train/yolov5m_csl_dotav1.5/weights`.
+
+1. Test yolov5-obb with single GPU. Get the HBB metrics.
+
+```shell
+python val.py \
+ --data 'data/yolov5obb_demo_split.yaml' \
+ --weights 'runs/train/yolov5m_csl_dotav1.5/weights/best.pt' \
+ --batch-size 2 --img 1024 --task 'val' --device 0 --save-json --name 'obb_demo_split'
+
+ Class Images Labels P R mAP@.5 mAP@.5:.95: 100%|██████████| 3/3 [00:02<00:00, 1.09it/s]
+ all 6 68 0.921 0.914 0.966 0.776
+ plane 6 16 0.946 1 0.995 0.934
+ small-vehicle 6 35 0.928 0.741 0.916 0.599
+ large-vehicle 6 17 0.89 1 0.986 0.793
+Speed: .................................................... per image at shape (2, 3, 1024, 1024)
+...
+Evaluating pycocotools mAP... saving runs/val/obb_demo_split/best_obb_predictions.json...
+---------------------The hbb and obb results has been saved in json file-----------------------
+```
+
+2. Parse the results. Get the poly format results.
+```shell
+python tools/TestJson2VocClassTxt.py --json_path 'runs/val/obb_demo_split/best_obb_predictions.json' --save_path 'runs/val/obb_demo_split/obb_predictions_Txt'
+```
+
+3. Merge the results. (If you split your dataset)
+```shell
+python DOTA_devkit/ResultMerge_multi_process.py \
+ --scrpath 'runs/val/obb_demo_split/obb_predictions_Txt' \
+ --dstpath 'runs/val/obb_demo_split/obb_predictions_Txt_Merged'
+```
+
+4. Get the OBB metrics
+```shell
+python DOTA_devkit/dota_evaluation_task1.py \
+ --detpath 'runs/val/obb_demo_split/obb_predictions_Txt_Merged/Task1_{:s}.txt' \
+ --annopath 'dataset/dataset_demo/labelTxt/{:s}.txt' \
+ --imagesetfile 'dataset/dataset_demo/imgnamefile.txt'
+
+...
+map: 0.6666666666666669
+classaps: [100. 0. 100.]
+```
+
+# Inferenece with pretrained models. (Original Dataset)
+We provide the validation/testing scripts to evaluate the trained model.
+
+Examples:
+
+Assume that you have already downloaded the checkpoints to `runs/train/yolov5m_csl_dotav1.5/weights`.
+
+1. Test yolov5-obb with single GPU. Get the HBB metrics.
+
+```shell
+python val.py \
+ --data 'data/yolov5obb_demo.yaml' \
+ --weights 'runs/train/yolov5m_csl_dotav1.5/weights/best.pt' \
+ --batch-size 1 --img 2048 --task 'val' --device 0 --save-json --name 'obb_demo'
+
+ Class Images Labels P R mAP@.5 mAP@.5:.95: 100%|██████████| 1/1 [00:00<00:00, 1.98it/s]
+ all 1 56 0.97 0.85 0.953 0.752
+ plane 1 11 1 1 0.995 0.944
+ small-vehicle 1 34 1 0.641 0.889 0.535
+ large-vehicle 1 11 0.91 0.909 0.976 0.777
+Speed: .................................................... per image at shape (1, 3, 2048, 2048)
+...
+Evaluating pycocotools mAP... saving runs/val/obb_demo/best_obb_predictions.json...
+---------------------The hbb and obb results has been saved in json file-----------------------
+```
+
+2. Parse the results. Get the poly format results.
+```shell
+python tools/TestJson2VocClassTxt.py --json_path 'runs/val/obb_demo/best_obb_predictions.json' --save_path 'runs/val/obb_demo/obb_predictions_Txt'
+```
+
+3. Get the OBB metrics
+```shell
+python DOTA_devkit/dota_evaluation_task1.py \
+ --detpath 'runs/val/obb_demo/obb_predictions_Txt/Task1_{:s}.txt' \
+ --annopath 'dataset/dataset_demo/labelTxt/{:s}.txt' \
+ --imagesetfile 'dataset/dataset_demo/imgnamefile.txt'
+
+...
+map: 0.6666666666666669
+classaps: [100. 0. 100.]
+```
+
+# Run inference on images, videos, directories, streams, etc. Then save the detection file.
+1. image demo
+```shell
+python detect.py --weights 'runs/train/yolov5m_csl_dotav1.5/weights/best.pt' \
+ --source 'dataset/dataset_demo/images/' \
+ --img 2048 --device 0 --conf-thres 0.25 --iou-thres 0.2 --hide-labels --hide-conf
+```
+
+***If you want to evaluate the result on DOTA test-dev, please zip the poly format results files and submit it to the [evaluation server](https://captain-whu.github.io/DOTA/index.html).**
diff --git a/docs/YOLOv5_README.md b/docs/YOLOv5_README.md
new file mode 100644
index 0000000..fa0645d
--- /dev/null
+++ b/docs/YOLOv5_README.md
@@ -0,0 +1,292 @@
+
+YOLOv5 🚀 is a family of object detection architectures and models pretrained on the COCO dataset, and represents Ultralytics
+ open-source research into future vision AI methods, incorporating lessons learned and best practices evolved over thousands of hours of research and development.
+
+
+
+
+
+
+##
Documentation
+
+See the [YOLOv5 Docs](https://docs.ultralytics.com) for full documentation on training, testing and deployment.
+
+##
Quick Start Examples
+
+
+Install
+
+[**Python>=3.6.0**](https://www.python.org/) is required with all
+[requirements.txt](https://github.com/ultralytics/yolov5/blob/master/requirements.txt) installed including
+[**PyTorch>=1.7**](https://pytorch.org/get-started/locally/):
+
+
+```bash
+$ git clone https://github.com/ultralytics/yolov5
+$ cd yolov5
+$ pip install -r requirements.txt
+```
+
+
+
+
+Inference
+
+Inference with YOLOv5 and [PyTorch Hub](https://github.com/ultralytics/yolov5/issues/36). Models automatically download
+from the [latest YOLOv5 release](https://github.com/ultralytics/yolov5/releases).
+
+```python
+import torch
+
+# Model
+model = torch.hub.load('ultralytics/yolov5', 'yolov5s') # or yolov5m, yolov5l, yolov5x, custom
+
+# Images
+img = 'https://ultralytics.com/images/zidane.jpg' # or file, Path, PIL, OpenCV, numpy, list
+
+# Inference
+results = model(img)
+
+# Results
+results.print() # or .show(), .save(), .crop(), .pandas(), etc.
+```
+
+
+
+
+
+
+Inference with detect.py
+
+`detect.py` runs inference on a variety of sources, downloading models automatically from
+the [latest YOLOv5 release](https://github.com/ultralytics/yolov5/releases) and saving results to `runs/detect`.
+
+```bash
+$ python detect.py --source 0 # webcam
+ img.jpg # image
+ vid.mp4 # video
+ path/ # directory
+ path/*.jpg # glob
+ 'https://youtu.be/Zgi9g1ksQHc' # YouTube
+ 'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP stream
+```
+
+
+
+
+Training
+
+Run commands below to reproduce results
+on [COCO](https://github.com/ultralytics/yolov5/blob/master/data/scripts/get_coco.sh) dataset (dataset auto-downloads on
+first use). Training times for YOLOv5s/m/l/x are 2/4/6/8 days on a single V100 (multi-GPU times faster). Use the
+largest `--batch-size` your GPU allows (batch sizes shown for 16 GB devices).
+
+```bash
+$ python train.py --data coco.yaml --cfg yolov5s.yaml --weights '' --batch-size 64
+ yolov5m 40
+ yolov5l 24
+ yolov5x 16
+```
+
+
+
+
+
+
+Tutorials
+
+* [Train Custom Data](https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data) 🚀 RECOMMENDED
+* [Tips for Best Training Results](https://github.com/ultralytics/yolov5/wiki/Tips-for-Best-Training-Results) ☘️
+ RECOMMENDED
+* [Weights & Biases Logging](https://github.com/ultralytics/yolov5/issues/1289) 🌟 NEW
+* [Roboflow for Datasets, Labeling, and Active Learning](https://github.com/ultralytics/yolov5/issues/4975) 🌟 NEW
+* [Multi-GPU Training](https://github.com/ultralytics/yolov5/issues/475)
+* [PyTorch Hub](https://github.com/ultralytics/yolov5/issues/36) ⭐ NEW
+* [TFLite, ONNX, CoreML, TensorRT Export](https://github.com/ultralytics/yolov5/issues/251) 🚀
+* [Test-Time Augmentation (TTA)](https://github.com/ultralytics/yolov5/issues/303)
+* [Model Ensembling](https://github.com/ultralytics/yolov5/issues/318)
+* [Model Pruning/Sparsity](https://github.com/ultralytics/yolov5/issues/304)
+* [Hyperparameter Evolution](https://github.com/ultralytics/yolov5/issues/607)
+* [Transfer Learning with Frozen Layers](https://github.com/ultralytics/yolov5/issues/1314) ⭐ NEW
+* [TensorRT Deployment](https://github.com/wang-xinyu/tensorrtx)
+
+
+
+##
Environments
+
+Get started in seconds with our verified environments. Click each icon below for details.
+
+
+
+|Weights and Biases|Roboflow ⭐ NEW|
+|:-:|:-:|
+|Automatically track and visualize all your YOLOv5 training runs in the cloud with [Weights & Biases](https://wandb.ai/site?utm_campaign=repo_yolo_readme)|Label and export your custom datasets directly to YOLOv5 for training with [Roboflow](https://roboflow.com/?ref=ultralytics) |
+
+
+
+
+##
Why YOLOv5
+
+
+
+ YOLOv5-P5 640 Figure (click to expand)
+
+
+
+
+ Figure Notes (click to expand)
+
+* **COCO AP val** denotes mAP@0.5:0.95 metric measured on the 5000-image [COCO val2017](http://cocodataset.org) dataset over various inference sizes from 256 to 1536.
+* **GPU Speed** measures average inference time per image on [COCO val2017](http://cocodataset.org) dataset using a [AWS p3.2xlarge](https://aws.amazon.com/ec2/instance-types/p3/) V100 instance at batch-size 32.
+* **EfficientDet** data from [google/automl](https://github.com/google/automl) at batch size 8.
+* **Reproduce** by `python val.py --task study --data coco.yaml --iou 0.7 --weights yolov5n6.pt yolov5s6.pt yolov5m6.pt yolov5l6.pt yolov5x6.pt`
+
+
+### Pretrained Checkpoints
+
+[assets]: https://github.com/ultralytics/yolov5/releases
+[TTA]: https://github.com/ultralytics/yolov5/issues/303
+
+|Model |size (pixels) |mAPval 0.5:0.95 |mAPval 0.5 |Speed CPU b1 (ms) |Speed V100 b1 (ms) |Speed V100 b32 (ms) |params (M) |FLOPs @640 (B)
+|--- |--- |--- |--- |--- |--- |--- |--- |---
+|[YOLOv5n][assets] |640 |28.4 |46.0 |**45** |**6.3**|**0.6**|**1.9**|**4.5**
+|[YOLOv5s][assets] |640 |37.2 |56.0 |98 |6.4 |0.9 |7.2 |16.5
+|[YOLOv5m][assets] |640 |45.2 |63.9 |224 |8.2 |1.7 |21.2 |49.0
+|[YOLOv5l][assets] |640 |48.8 |67.2 |430 |10.1 |2.7 |46.5 |109.1
+|[YOLOv5x][assets] |640 |50.7 |68.9 |766 |12.1 |4.8 |86.7 |205.7
+| | | | | | | | |
+|[YOLOv5n6][assets] |1280 |34.0 |50.7 |153 |8.1 |2.1 |3.2 |4.6
+|[YOLOv5s6][assets] |1280 |44.5 |63.0 |385 |8.2 |3.6 |12.6 |16.8
+|[YOLOv5m6][assets] |1280 |51.0 |69.0 |887 |11.1 |6.8 |35.7 |50.0
+|[YOLOv5l6][assets] |1280 |53.6 |71.6 |1784 |15.8 |10.5 |76.7 |111.4
+|[YOLOv5x6][assets] + [TTA][TTA]|1280 1536 |54.7 **55.4** |**72.4** 72.3 |3136 - |26.2 - |19.4 - |140.7 - |209.8 -
+
+
+ Table Notes (click to expand)
+
+* All checkpoints are trained to 300 epochs with default settings and hyperparameters.
+* **mAPval** values are for single-model single-scale on [COCO val2017](http://cocodataset.org) dataset. Reproduce by `python val.py --data coco.yaml --img 640 --conf 0.001 --iou 0.65`
+* **Speed** averaged over COCO val images using a [AWS p3.2xlarge](https://aws.amazon.com/ec2/instance-types/p3/) instance. NMS times (~1 ms/img) not included. Reproduce by `python val.py --data coco.yaml --img 640 --conf 0.25 --iou 0.45`
+* **TTA** [Test Time Augmentation](https://github.com/ultralytics/yolov5/issues/303) includes reflection and scale augmentations. Reproduce by `python val.py --data coco.yaml --img 1536 --iou 0.7 --augment`
+
+
+
+##
Contribute
+
+We love your input! We want to make contributing to YOLOv5 as easy and transparent as possible. Please see our [Contributing Guide](CONTRIBUTING.md) to get started, and fill out the [YOLOv5 Survey](https://ultralytics.com/survey?utm_source=github&utm_medium=social&utm_campaign=Survey) to send us feedback on your experiences. Thank you to all our contributors!
+
+
+
+
+##
Contact
+
+For YOLOv5 bugs and feature requests please visit [GitHub Issues](https://github.com/ultralytics/yolov5/issues). For business inquiries or
+professional support requests please visit [https://ultralytics.com/contact](https://ultralytics.com/contact).
+
+
+
+
\n",
+ "Close the active learning loop by sampling images from your inference conditions with the `roboflow` pip package\n",
+ "
\n",
+ "\n",
+ "Train a YOLOv5s model on the [COCO128](https://www.kaggle.com/ultralytics/coco128) dataset with `--data coco128.yaml`, starting from pretrained `--weights yolov5s.pt`, or from randomly initialized `--weights '' --cfg yolov5s.yaml`.\n",
+ "\n",
+ "- **Pretrained [Models](https://github.com/ultralytics/yolov5/tree/master/models)** are downloaded\n",
+ "automatically from the [latest YOLOv5 release](https://github.com/ultralytics/yolov5/releases)\n",
+ "- **[Datasets](https://github.com/ultralytics/yolov5/tree/master/data)** available for autodownload include: [COCO](https://github.com/ultralytics/yolov5/blob/master/data/coco.yaml), [COCO128](https://github.com/ultralytics/yolov5/blob/master/data/coco128.yaml), [VOC](https://github.com/ultralytics/yolov5/blob/master/data/VOC.yaml), [Argoverse](https://github.com/ultralytics/yolov5/blob/master/data/Argoverse.yaml), [VisDrone](https://github.com/ultralytics/yolov5/blob/master/data/VisDrone.yaml), [GlobalWheat](https://github.com/ultralytics/yolov5/blob/master/data/GlobalWheat2020.yaml), [xView](https://github.com/ultralytics/yolov5/blob/master/data/xView.yaml), [Objects365](https://github.com/ultralytics/yolov5/blob/master/data/Objects365.yaml), [SKU-110K](https://github.com/ultralytics/yolov5/blob/master/data/SKU-110K.yaml).\n",
+ "- **Training Results** are saved to `runs/train/` with incrementing run directories, i.e. `runs/train/exp2`, `runs/train/exp3` etc.\n",
+ "
\n",
+ "\n",
+ "## Train on Custom Data with Roboflow 🌟 NEW\n",
+ "\n",
+ "[Roboflow](https://roboflow.com/?ref=ultralytics) enables you to easily **organize, label, and prepare** a high quality dataset with your own custom data. Roboflow also makes it easy to establish an active learning pipeline, collaborate with your team on dataset improvement, and integrate directly into your model building workflow with the `roboflow` pip package.\n",
+ "\n",
+ "- Custom Training Example: [https://blog.roboflow.com/how-to-train-yolov5-on-a-custom-dataset/](https://blog.roboflow.com/how-to-train-yolov5-on-a-custom-dataset/?ref=ultralytics)\n",
+ "- Custom Training Notebook: [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/roboflow-ai/yolov5-custom-training-tutorial/blob/main/yolov5-custom-training.ipynb)\n",
+ " \n",
+ "\n",
+ "
+
+
+
\ No newline at end of file
diff --git a/utils/.idea/inspectionProfiles/profiles_settings.xml b/utils/.idea/inspectionProfiles/profiles_settings.xml
new file mode 100644
index 0000000..105ce2d
--- /dev/null
+++ b/utils/.idea/inspectionProfiles/profiles_settings.xml
@@ -0,0 +1,6 @@
+
+
+
+
+
+
\ No newline at end of file
diff --git a/utils/.idea/misc.xml b/utils/.idea/misc.xml
new file mode 100644
index 0000000..4e1828e
--- /dev/null
+++ b/utils/.idea/misc.xml
@@ -0,0 +1,4 @@
+
+
+
+
\ No newline at end of file
diff --git a/utils/.idea/modules.xml b/utils/.idea/modules.xml
new file mode 100644
index 0000000..2213c52
--- /dev/null
+++ b/utils/.idea/modules.xml
@@ -0,0 +1,8 @@
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/utils/.idea/utils.iml b/utils/.idea/utils.iml
new file mode 100644
index 0000000..a1e094c
--- /dev/null
+++ b/utils/.idea/utils.iml
@@ -0,0 +1,12 @@
+
+
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/utils/__init__.py b/utils/__init__.py
new file mode 100644
index 0000000..295aebf
--- /dev/null
+++ b/utils/__init__.py
@@ -0,0 +1,37 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+utils/initialization
+"""
+
+
+def notebook_init(verbose=True):
+ # Check system software and hardware
+ print('Checking setup...')
+
+ import os
+ import shutil
+
+ from utils.general import check_requirements, emojis, is_colab
+ from utils.torch_utils import select_device # imports
+
+ check_requirements(('psutil', 'IPython'))
+ import psutil
+ from IPython import display # to display images and clear console output
+
+ if is_colab():
+ shutil.rmtree('/content/sample_data', ignore_errors=True) # remove colab /sample_data directory
+
+ if verbose:
+ # System info
+ # gb = 1 / 1000 ** 3 # bytes to GB
+ gib = 1 / 1024 ** 3 # bytes to GiB
+ ram = psutil.virtual_memory().total
+ total, used, free = shutil.disk_usage("/")
+ display.clear_output()
+ s = f'({os.cpu_count()} CPUs, {ram * gib:.1f} GB RAM, {(total - free) * gib:.1f}/{total * gib:.1f} GB disk)'
+ else:
+ s = ''
+
+ select_device(newline=False)
+ print(emojis(f'Setup complete ✅ {s}'))
+ return display
diff --git a/utils/__pycache__/__init__.cpython-37.pyc b/utils/__pycache__/__init__.cpython-37.pyc
new file mode 100644
index 0000000..8a05961
Binary files /dev/null and b/utils/__pycache__/__init__.cpython-37.pyc differ
diff --git a/utils/__pycache__/augmentations.cpython-37.pyc b/utils/__pycache__/augmentations.cpython-37.pyc
new file mode 100644
index 0000000..9b60481
Binary files /dev/null and b/utils/__pycache__/augmentations.cpython-37.pyc differ
diff --git a/utils/__pycache__/autoanchor.cpython-37.pyc b/utils/__pycache__/autoanchor.cpython-37.pyc
new file mode 100644
index 0000000..5dba0d7
Binary files /dev/null and b/utils/__pycache__/autoanchor.cpython-37.pyc differ
diff --git a/utils/__pycache__/autobatch.cpython-37.pyc b/utils/__pycache__/autobatch.cpython-37.pyc
new file mode 100644
index 0000000..f710af8
Binary files /dev/null and b/utils/__pycache__/autobatch.cpython-37.pyc differ
diff --git a/utils/__pycache__/callbacks.cpython-37.pyc b/utils/__pycache__/callbacks.cpython-37.pyc
new file mode 100644
index 0000000..748cdbb
Binary files /dev/null and b/utils/__pycache__/callbacks.cpython-37.pyc differ
diff --git a/utils/__pycache__/datasets.cpython-37.pyc b/utils/__pycache__/datasets.cpython-37.pyc
new file mode 100644
index 0000000..2b41887
Binary files /dev/null and b/utils/__pycache__/datasets.cpython-37.pyc differ
diff --git a/utils/__pycache__/downloads.cpython-37.pyc b/utils/__pycache__/downloads.cpython-37.pyc
new file mode 100644
index 0000000..25e35c0
Binary files /dev/null and b/utils/__pycache__/downloads.cpython-37.pyc differ
diff --git a/utils/__pycache__/general.cpython-37.pyc b/utils/__pycache__/general.cpython-37.pyc
new file mode 100644
index 0000000..b8e613e
Binary files /dev/null and b/utils/__pycache__/general.cpython-37.pyc differ
diff --git a/utils/__pycache__/loss.cpython-37.pyc b/utils/__pycache__/loss.cpython-37.pyc
new file mode 100644
index 0000000..e71e0ac
Binary files /dev/null and b/utils/__pycache__/loss.cpython-37.pyc differ
diff --git a/utils/__pycache__/metrics.cpython-37.pyc b/utils/__pycache__/metrics.cpython-37.pyc
new file mode 100644
index 0000000..69af152
Binary files /dev/null and b/utils/__pycache__/metrics.cpython-37.pyc differ
diff --git a/utils/__pycache__/plots.cpython-37.pyc b/utils/__pycache__/plots.cpython-37.pyc
new file mode 100644
index 0000000..e9a5c69
Binary files /dev/null and b/utils/__pycache__/plots.cpython-37.pyc differ
diff --git a/utils/__pycache__/rboxs_utils.cpython-37.pyc b/utils/__pycache__/rboxs_utils.cpython-37.pyc
new file mode 100644
index 0000000..bf73be2
Binary files /dev/null and b/utils/__pycache__/rboxs_utils.cpython-37.pyc differ
diff --git a/utils/__pycache__/torch_utils.cpython-37.pyc b/utils/__pycache__/torch_utils.cpython-37.pyc
new file mode 100644
index 0000000..c4f9c5c
Binary files /dev/null and b/utils/__pycache__/torch_utils.cpython-37.pyc differ
diff --git a/utils/activations.py b/utils/activations.py
new file mode 100644
index 0000000..a4ff789
--- /dev/null
+++ b/utils/activations.py
@@ -0,0 +1,101 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Activation functions
+"""
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+# SiLU https://arxiv.org/pdf/1606.08415.pdf ----------------------------------------------------------------------------
+class SiLU(nn.Module): # export-friendly version of nn.SiLU()
+ @staticmethod
+ def forward(x):
+ return x * torch.sigmoid(x)
+
+
+class Hardswish(nn.Module): # export-friendly version of nn.Hardswish()
+ @staticmethod
+ def forward(x):
+ # return x * F.hardsigmoid(x) # for TorchScript and CoreML
+ return x * F.hardtanh(x + 3, 0.0, 6.0) / 6.0 # for TorchScript, CoreML and ONNX
+
+
+# Mish https://github.com/digantamisra98/Mish --------------------------------------------------------------------------
+class Mish(nn.Module):
+ @staticmethod
+ def forward(x):
+ return x * F.softplus(x).tanh()
+
+
+class MemoryEfficientMish(nn.Module):
+ class F(torch.autograd.Function):
+ @staticmethod
+ def forward(ctx, x):
+ ctx.save_for_backward(x)
+ return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x)))
+
+ @staticmethod
+ def backward(ctx, grad_output):
+ x = ctx.saved_tensors[0]
+ sx = torch.sigmoid(x)
+ fx = F.softplus(x).tanh()
+ return grad_output * (fx + x * sx * (1 - fx * fx))
+
+ def forward(self, x):
+ return self.F.apply(x)
+
+
+# FReLU https://arxiv.org/abs/2007.11824 -------------------------------------------------------------------------------
+class FReLU(nn.Module):
+ def __init__(self, c1, k=3): # ch_in, kernel
+ super().__init__()
+ self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1, bias=False)
+ self.bn = nn.BatchNorm2d(c1)
+
+ def forward(self, x):
+ return torch.max(x, self.bn(self.conv(x)))
+
+
+# ACON https://arxiv.org/pdf/2009.04759.pdf ----------------------------------------------------------------------------
+class AconC(nn.Module):
+ r""" ACON activation (activate or not).
+ AconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is a learnable parameter
+ according to "Activate or Not: Learning Customized Activation" .
+ """
+
+ def __init__(self, c1):
+ super().__init__()
+ self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
+ self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
+ self.beta = nn.Parameter(torch.ones(1, c1, 1, 1))
+
+ def forward(self, x):
+ dpx = (self.p1 - self.p2) * x
+ return dpx * torch.sigmoid(self.beta * dpx) + self.p2 * x
+
+
+class MetaAconC(nn.Module):
+ r""" ACON activation (activate or not).
+ MetaAconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is generated by a small network
+ according to "Activate or Not: Learning Customized Activation" .
+ """
+
+ def __init__(self, c1, k=1, s=1, r=16): # ch_in, kernel, stride, r
+ super().__init__()
+ c2 = max(r, c1 // r)
+ self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
+ self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
+ self.fc1 = nn.Conv2d(c1, c2, k, s, bias=True)
+ self.fc2 = nn.Conv2d(c2, c1, k, s, bias=True)
+ # self.bn1 = nn.BatchNorm2d(c2)
+ # self.bn2 = nn.BatchNorm2d(c1)
+
+ def forward(self, x):
+ y = x.mean(dim=2, keepdims=True).mean(dim=3, keepdims=True)
+ # batch-size 1 bug/instabilities https://github.com/ultralytics/yolov5/issues/2891
+ # beta = torch.sigmoid(self.bn2(self.fc2(self.bn1(self.fc1(y))))) # bug/unstable
+ beta = torch.sigmoid(self.fc2(self.fc1(y))) # bug patch BN layers removed
+ dpx = (self.p1 - self.p2) * x
+ return dpx * torch.sigmoid(beta * dpx) + self.p2 * x
diff --git a/utils/augmentations.py b/utils/augmentations.py
new file mode 100644
index 0000000..53ccc9a
--- /dev/null
+++ b/utils/augmentations.py
@@ -0,0 +1,289 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Image augmentation functions
+"""
+
+import math
+import random
+
+import cv2
+import numpy as np
+
+from utils.general import LOGGER, check_version, colorstr, resample_segments, segment2box
+from utils.metrics import bbox_ioa
+from utils.rboxs_utils import poly_filter
+
+
+class Albumentations:
+ # YOLOv5 Albumentations class (optional, only used if package is installed)
+ def __init__(self):
+ self.transform = None
+ try:
+ import albumentations as A
+ check_version(A.__version__, '1.0.3', hard=True) # version requirement
+
+ self.transform = A.Compose([
+ A.Blur(p=0.01),
+ A.MedianBlur(p=0.01),
+ A.ToGray(p=0.01),
+ A.CLAHE(p=0.01),
+ A.RandomBrightnessContrast(p=0.0),
+ A.RandomGamma(p=0.0),
+ A.ImageCompression(quality_lower=75, p=0.0)],
+ bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))
+
+ LOGGER.info(colorstr('albumentations: ') + ', '.join(f'{x}' for x in self.transform.transforms if x.p))
+ except ImportError: # package not installed, skip
+ pass
+ except Exception as e:
+ LOGGER.info(colorstr('albumentations: ') + f'{e}')
+
+ def __call__(self, im, labels, p=1.0):
+ if self.transform and random.random() < p:
+ new = self.transform(image=im, bboxes=labels[:, 1:], class_labels=labels[:, 0]) # transformed
+ im, labels = new['image'], np.array([[c, *b] for c, b in zip(new['class_labels'], new['bboxes'])])
+ return im, labels
+
+
+def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5):
+ # HSV color-space augmentation
+ if hgain or sgain or vgain:
+ r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1 # random gains
+ hue, sat, val = cv2.split(cv2.cvtColor(im, cv2.COLOR_BGR2HSV))
+ dtype = im.dtype # uint8
+
+ x = np.arange(0, 256, dtype=r.dtype)
+ lut_hue = ((x * r[0]) % 180).astype(dtype)
+ lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
+ lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
+
+ im_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val)))
+ cv2.cvtColor(im_hsv, cv2.COLOR_HSV2BGR, dst=im) # no return needed
+
+
+def hist_equalize(im, clahe=True, bgr=False):
+ # Equalize histogram on BGR image 'im' with im.shape(n,m,3) and range 0-255
+ yuv = cv2.cvtColor(im, cv2.COLOR_BGR2YUV if bgr else cv2.COLOR_RGB2YUV)
+ if clahe:
+ c = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
+ yuv[:, :, 0] = c.apply(yuv[:, :, 0])
+ else:
+ yuv[:, :, 0] = cv2.equalizeHist(yuv[:, :, 0]) # equalize Y channel histogram
+ return cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR if bgr else cv2.COLOR_YUV2RGB) # convert YUV image to RGB
+
+
+def replicate(im, labels):
+ # Replicate labels
+ h, w = im.shape[:2]
+ boxes = labels[:, 1:].astype(int)
+ x1, y1, x2, y2 = boxes.T
+ s = ((x2 - x1) + (y2 - y1)) / 2 # side length (pixels)
+ for i in s.argsort()[:round(s.size * 0.5)]: # smallest indices
+ x1b, y1b, x2b, y2b = boxes[i]
+ bh, bw = y2b - y1b, x2b - x1b
+ yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw)) # offset x, y
+ x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
+ im[y1a:y2a, x1a:x2a] = im[y1b:y2b, x1b:x2b] # im4[ymin:ymax, xmin:xmax]
+ labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)
+
+ return im, labels
+
+
+def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
+ """
+ Resize and pad image while meeting stride-multiple constraints
+ Returns:
+ im (array): (height, width, 3)
+ ratio (array): [w_ratio, h_ratio]
+ (dw, dh) (array): [w_padding h_padding]
+ """
+ shape = im.shape[:2] # current shape [height, width]
+ if isinstance(new_shape, int): # [h_rect, w_rect]
+ new_shape = (new_shape, new_shape)
+
+ # Scale ratio (new / old)
+ r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
+ if not scaleup: # only scale down, do not scale up (for better val mAP)
+ r = min(r, 1.0)
+
+ # Compute padding
+ ratio = r, r # wh ratios
+ new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r)) # w h
+ dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1] # wh padding
+ if auto: # minimum rectangle
+ dw, dh = np.mod(dw, stride), np.mod(dh, stride) # wh padding
+ elif scaleFill: # stretch
+ dw, dh = 0.0, 0.0
+ new_unpad = (new_shape[1], new_shape[0]) # [w h]
+ ratio = new_shape[1] / shape[1], new_shape[0] / shape[0] # [w_ratio, h_ratio]
+
+ dw /= 2 # divide padding into 2 sides
+ dh /= 2
+
+ if shape[::-1] != new_unpad: # resize
+ im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
+ top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
+ left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
+ im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color) # add border
+ return im, ratio, (dw, dh)
+
+
+def random_perspective(im, targets=(), segments=(), degrees=10, translate=.1, scale=.1, shear=10, perspective=0.0,
+ border=(0, 0)):
+ # torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(0.1, 0.1), scale=(0.9, 1.1), shear=(-10, 10))
+ # targets = [cls, xyxyxyxy]
+
+ height = im.shape[0] + border[0] * 2 # shape(h,w,c)
+ width = im.shape[1] + border[1] * 2
+
+ # Center
+ C = np.eye(3)
+ C[0, 2] = -im.shape[1] / 2 # x translation (pixels)
+ C[1, 2] = -im.shape[0] / 2 # y translation (pixels)
+
+ # Perspective
+ P = np.eye(3)
+ P[2, 0] = random.uniform(-perspective, perspective) # x perspective (about y)
+ P[2, 1] = random.uniform(-perspective, perspective) # y perspective (about x)
+
+ # Rotation and Scale
+ R = np.eye(3)
+ a = random.uniform(-degrees, degrees)
+ # a += random.choice([-180, -90, 0, 90]) # add 90deg rotations to small rotations
+ s = random.uniform(1 - scale, 1 + scale)
+ # s = 2 ** random.uniform(-scale, scale)
+ R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
+
+ # Shear
+ S = np.eye(3)
+ S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # x shear (deg)
+ S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # y shear (deg)
+
+ # Translation
+ T = np.eye(3)
+ T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width # x translation (pixels)
+ T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height # y translation (pixels)
+
+ # Combined rotation matrix
+ M = T @ S @ R @ P @ C # order of operations (right to left) is IMPORTANT
+ if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any(): # image changed
+ if perspective:
+ im = cv2.warpPerspective(im, M, dsize=(width, height), borderValue=(114, 114, 114))
+ else: # affine
+ im = cv2.warpAffine(im, M[:2], dsize=(width, height), borderValue=(114, 114, 114))
+
+ # Visualize
+ # import matplotlib.pyplot as plt
+ # ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
+ # ax[0].imshow(im[:, :, ::-1]) # base
+ # ax[1].imshow(im2[:, :, ::-1]) # warped
+
+ # Transform label coordinates
+ n = len(targets)
+ if n:
+ use_segments = any(x.any() for x in segments)
+ new = np.zeros((n, 4))
+ if use_segments: # warp segments
+ segments = resample_segments(segments) # upsample
+ for i, segment in enumerate(segments):
+ xy = np.ones((len(segment), 3))
+ xy[:, :2] = segment
+ xy = xy @ M.T # transform
+ xy = xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2] # perspective rescale or affine
+
+ # clip
+ new[i] = segment2box(xy, width, height)
+
+ else: # warp boxes
+ xy = np.ones((n * 4, 3))
+ # xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2) # x1y1, x2y2, x1y2, x2y1
+ xy[:, :2] = targets[:, 1:].reshape(n * 4, 2) # x1y1, x2y2, x1y2, x2y1
+ xy = xy @ M.T # transform
+ xy = (xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]).reshape(n, 8) # perspective rescale or affine
+
+ # # create new boxes
+ # x = xy[:, [0, 2, 4, 6]]
+ # y = xy[:, [1, 3, 5, 7]]
+ # new = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T
+
+ # # clip
+ # new[:, [0, 2]] = new[:, [0, 2]].clip(0, width)
+ # new[:, [1, 3]] = new[:, [1, 3]].clip(0, height)
+ # clip boxes 不启用,保留预测完整物体的能力
+
+ # filter candidates
+ # i = box_candidates(box1=targets[:, 1:5].T * s, box2=new.T, area_thr=0.01 if use_segments else 0.10)
+ # targets = targets[i]
+ # targets[:, 1:5] = new[i]
+ targets_mask = poly_filter(polys=xy, h=height, w=width)
+ targets[:, 1:] = xy
+ targets = targets[targets_mask]
+
+ return im, targets
+
+
+def copy_paste(im, labels, segments, p=0.5):
+ # Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy)
+ n = len(segments)
+ if p and n:
+ h, w, c = im.shape # height, width, channels
+ im_new = np.zeros(im.shape, np.uint8)
+ for j in random.sample(range(n), k=round(p * n)):
+ l, s = labels[j], segments[j]
+ box = w - l[3], l[2], w - l[1], l[4]
+ ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area
+ if (ioa < 0.30).all(): # allow 30% obscuration of existing labels
+ labels = np.concatenate((labels, [[l[0], *box]]), 0)
+ segments.append(np.concatenate((w - s[:, 0:1], s[:, 1:2]), 1))
+ cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED)
+
+ result = cv2.bitwise_and(src1=im, src2=im_new)
+ result = cv2.flip(result, 1) # augment segments (flip left-right)
+ i = result > 0 # pixels to replace
+ # i[:, :] = result.max(2).reshape(h, w, 1) # act over ch
+ im[i] = result[i] # cv2.imwrite('debug.jpg', im) # debug
+
+ return im, labels, segments
+
+
+def cutout(im, labels, p=0.5):
+ # Applies image cutout augmentation https://arxiv.org/abs/1708.04552
+ if random.random() < p:
+ h, w = im.shape[:2]
+ scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16 # image size fraction
+ for s in scales:
+ mask_h = random.randint(1, int(h * s)) # create random masks
+ mask_w = random.randint(1, int(w * s))
+
+ # box
+ xmin = max(0, random.randint(0, w) - mask_w // 2)
+ ymin = max(0, random.randint(0, h) - mask_h // 2)
+ xmax = min(w, xmin + mask_w)
+ ymax = min(h, ymin + mask_h)
+
+ # apply random color mask
+ im[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)]
+
+ # return unobscured labels
+ if len(labels) and s > 0.03:
+ box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
+ ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area
+ labels = labels[ioa < 0.60] # remove >60% obscured labels
+
+ return labels
+
+
+def mixup(im, labels, im2, labels2):
+ # Applies MixUp augmentation https://arxiv.org/pdf/1710.09412.pdf
+ r = np.random.beta(32.0, 32.0) # mixup ratio, alpha=beta=32.0
+ im = (im * r + im2 * (1 - r)).astype(np.uint8)
+ labels = np.concatenate((labels, labels2), 0)
+ return im, labels
+
+
+def box_candidates(box1, box2, wh_thr=2, ar_thr=100, area_thr=0.1, eps=1e-16): # box1(4,n), box2(4,n)
+ # Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
+ w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
+ w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
+ ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps)) # aspect ratio
+ return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr) # candidates
diff --git a/utils/autoanchor.py b/utils/autoanchor.py
new file mode 100644
index 0000000..6332113
--- /dev/null
+++ b/utils/autoanchor.py
@@ -0,0 +1,197 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Auto-anchor utils
+"""
+
+import random
+
+import numpy as np
+import torch
+import yaml
+from tqdm import tqdm
+
+from utils.general import LOGGER, colorstr, emojis
+from utils.rboxs_utils import pi, poly2rbox, regular_theta
+import cv2
+
+PREFIX = colorstr('AutoAnchor: ')
+
+
+def check_anchor_order(m):
+ # Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary
+ a = m.anchors.prod(-1).view(-1) # anchor area
+ da = a[-1] - a[0] # delta a
+ ds = m.stride[-1] - m.stride[0] # delta s
+ if da.sign() != ds.sign(): # same order
+ LOGGER.info(f'{PREFIX}Reversing anchor order')
+ m.anchors[:] = m.anchors.flip(0)
+
+
+def check_anchors(dataset, model, thr=4.0, imgsz=640):
+ """
+ Args:
+ Dataset.labels (list): n_imgs * array(num_gt_perimg, [cls_id, poly])
+ Dataset.shapes (array): (n_imgs, [ori_img_width, ori_img_height])
+ Returns:
+
+ """
+ # Check anchor fit to data, recompute if necessary
+ m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect()
+ # shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True)
+ min_ratios = imgsz / dataset.shapes.max(1, keepdims=True) #
+ scales = np.random.uniform(0.9, 1.1, size=(min_ratios.shape[0], 1)) # augment scale
+
+ # wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float() # wh
+ ls_edges = []
+ for ratio, labels in zip(min_ratios * scales, dataset.labels): # labels (array): (num_gt_perimg, [cls_id, poly])
+ rboxes = poly2rbox(labels[:, 1:] * ratio)
+ if len(rboxes):
+ ls_edges.append(rboxes[:, 2:4])
+ ls_edges = torch.tensor(np.concatenate(ls_edges)).float()
+ ls_edges = ls_edges[(ls_edges >= 5.0).any(1)] # filter > 5 pixels, anchor 宽高不能都小于5
+
+ def metric(k): # compute metric
+ r = ls_edges[:, None] / k[None]
+ x = torch.min(r, 1 / r).min(2)[0] # ratio metric
+ best = x.max(1)[0] # best_x
+ aat = (x > 1 / thr).float().sum(1).mean() # anchors above threshold
+ bpr = (best > 1 / thr).float().mean() # best possible recall
+ return bpr, aat
+
+ anchors = m.anchors.clone() * m.stride.to(m.anchors.device).view(-1, 1, 1) # current anchors
+ bpr, aat = metric(anchors.cpu().view(-1, 2))
+ s = f'\n{PREFIX}{aat:.2f} anchors/target, {bpr:.3f} Best Possible Recall (BPR). '
+ if bpr > 0.98: # threshold to recompute
+ LOGGER.info(emojis(f'{s}Current anchors are a good fit to dataset ✅'))
+ else:
+ LOGGER.info(emojis(f'{s}Anchors are a poor fit to dataset ⚠️, attempting to improve...'))
+ na = m.anchors.numel() // 2 # number of anchors
+ try:
+ anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
+ except Exception as e:
+ LOGGER.info(f'{PREFIX}ERROR: {e}')
+ new_bpr = metric(anchors)[0]
+ if new_bpr > bpr: # replace anchors
+ anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors)
+ m.anchors[:] = anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1) # loss, featuremap stride pixel
+ check_anchor_order(m)
+ LOGGER.info(f'{PREFIX}New anchors saved to model. Update model *.yaml to use these anchors in the future.')
+ else:
+ LOGGER.info(f'{PREFIX}Original anchors better than new anchors. Proceeding with original anchors.')
+
+
+def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
+ """ Creates kmeans-evolved anchors from training dataset
+
+ Arguments:
+ dataset: path to data.yaml, or a loaded dataset
+ n: number of anchors
+ img_size: image size used for training
+ thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0
+ gen: generations to evolve anchors using genetic algorithm
+ verbose: print all results
+
+ Return:
+ k: kmeans evolved anchors
+
+ Usage:
+ from utils.autoanchor import *; _ = kmean_anchors()
+ """
+ from scipy.cluster.vq import kmeans
+
+ thr = 1 / thr
+
+ def metric(k, wh): # compute metrics
+ r = wh[:, None] / k[None]
+ x = torch.min(r, 1 / r).min(2)[0] # ratio metric
+ # x = wh_iou(wh, torch.tensor(k)) # iou metric
+ return x, x.max(1)[0] # x, best_x
+
+ def anchor_fitness(k): # mutation fitness
+ # _, best = metric(torch.tensor(k, dtype=torch.float32), wh)
+ _, best = metric(torch.tensor(k, dtype=torch.float32), ls_edges)
+ return (best * (best > thr).float()).mean() # fitness
+
+ def print_results(k, verbose=True):
+ k = k[np.argsort(k.prod(1))] # sort small to large
+ # x, best = metric(k, wh0)
+ x, best = metric(k, ls_edges0)
+ bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n # best possible recall, anch > thr
+ s = f'{PREFIX}thr={thr:.2f}: {bpr:.4f} best possible recall, {aat:.2f} anchors past thr\n' \
+ f'{PREFIX}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, ' \
+ f'past_thr={x[x > thr].mean():.3f}-mean: '
+ for i, x in enumerate(k):
+ s += '%i,%i, ' % (round(x[0]), round(x[1]))
+ if verbose:
+ LOGGER.info(s[:-2])
+ return k
+
+ if isinstance(dataset, str): # *.yaml file
+ with open(dataset, errors='ignore') as f:
+ data_dict = yaml.safe_load(f) # model dict
+ from utils.datasets import LoadImagesAndLabels
+ dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True)
+
+ # Get label l s
+ # shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True)
+ # wh0 = np.concatenate([l[:, 3:5] * s for s, l in zip(shapes, dataset.labels)]) # wh
+ min_ratios = img_size / dataset.shapes.max(1, keepdims=True) #
+ ls_edges0 = []
+ for ratio, labels in zip(min_ratios, dataset.labels): # labels (array): (num_gt_perimg, [cls_id, poly])
+ rboxes = poly2rbox(labels[:, 1:] * ratio)
+ if len(rboxes):
+ ls_edges0.append(rboxes[:, 2:4])
+ ls_edges0 = np.concatenate(ls_edges0)
+
+ # Filter
+ i = (ls_edges0 < 5.0).any(1).sum()
+ if i:
+ LOGGER.info(f'{PREFIX}WARNING: Extremely small objects found. {i} of {len(ls_edges0)} poly labels are < 5 pixels in size.')
+ # wh = wh0[(wh0 >= 2.0).any(1)] # filter > 2 pixels
+ ls_edges = ls_edges0[(ls_edges0 >= 5.0).any(1)] # filter > 5 pixels
+ # wh = wh * (np.random.rand(wh.shape[0], 1) * 0.9 + 0.1) # multiply by random scale 0-1
+
+ # Kmeans calculation
+ # LOGGER.info(f'{PREFIX}Running kmeans for {n} anchors on {len(wh)} points...')
+ # s = wh.std(0) # sigmas for whitening
+ # k, dist = kmeans(wh / s, n, iter=30) # points, mean distance
+ LOGGER.info(f'{PREFIX}Running kmeans for {n} anchors on {len(ls_edges)} points...')
+ s = ls_edges.std(0) # sigmas for whitening
+ k, dist = kmeans(ls_edges / s, n, iter=30) # points, mean distance
+ assert len(k) == n, f'{PREFIX}ERROR: scipy.cluster.vq.kmeans requested {n} points but returned only {len(k)}'
+ k *= s
+ # wh = torch.tensor(wh, dtype=torch.float32) # filtered
+ # wh0 = torch.tensor(wh0, dtype=torch.float32) # unfiltered
+ ls_edges = torch.tensor(ls_edges, dtype=torch.float32) # filtered
+ ls_edges0 = torch.tensor(ls_edges0, dtype=torch.float32) # unfiltered
+ k = print_results(k, verbose=False)
+
+ # Plot
+ # k, d = [None] * 20, [None] * 20
+ # for i in tqdm(range(1, 21)):
+ # k[i-1], d[i-1] = kmeans(wh / s, i) # points, mean distance
+ # fig, ax = plt.subplots(1, 2, figsize=(14, 7), tight_layout=True)
+ # ax = ax.ravel()
+ # ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.')
+ # fig, ax = plt.subplots(1, 2, figsize=(14, 7)) # plot wh
+ # ax[0].hist(wh[wh[:, 0]<100, 0],400)
+ # ax[1].hist(wh[wh[:, 1]<100, 1],400)
+ # fig.savefig('wh.png', dpi=200)
+
+ # Evolve
+ npr = np.random
+ f, sh, mp, s = anchor_fitness(k), k.shape, 0.9, 0.1 # fitness, generations, mutation prob, sigma
+ pbar = tqdm(range(gen), desc=f'{PREFIX}Evolving anchors with Genetic Algorithm:') # progress bar
+ for _ in pbar:
+ v = np.ones(sh)
+ while (v == 1).all(): # mutate until a change occurs (prevent duplicates)
+ v = ((npr.random(sh) < mp) * random.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0)
+ kg = (k.copy() * v).clip(min=2.0)
+ fg = anchor_fitness(kg)
+ if fg > f:
+ f, k = fg, kg.copy()
+ pbar.desc = f'{PREFIX}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}'
+ if verbose:
+ print_results(k, verbose)
+
+ return print_results(k)
diff --git a/utils/autobatch.py b/utils/autobatch.py
new file mode 100644
index 0000000..cb94f04
--- /dev/null
+++ b/utils/autobatch.py
@@ -0,0 +1,57 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Auto-batch utils
+"""
+
+from copy import deepcopy
+
+import numpy as np
+import torch
+from torch.cuda import amp
+
+from utils.general import LOGGER, colorstr
+from utils.torch_utils import profile
+
+
+def check_train_batch_size(model, imgsz=640):
+ # Check YOLOv5 training batch size
+ with amp.autocast():
+ return autobatch(deepcopy(model).train(), imgsz) # compute optimal batch size
+
+
+def autobatch(model, imgsz=640, fraction=0.9, batch_size=16):
+ # Automatically estimate best batch size to use `fraction` of available CUDA memory
+ # Usage:
+ # import torch
+ # from utils.autobatch import autobatch
+ # model = torch.hub.load('ultralytics/yolov5', 'yolov5s', autoshape=False)
+ # print(autobatch(model))
+
+ prefix = colorstr('AutoBatch: ')
+ LOGGER.info(f'{prefix}Computing optimal batch size for --imgsz {imgsz}')
+ device = next(model.parameters()).device # get model device
+ if device.type == 'cpu':
+ LOGGER.info(f'{prefix}CUDA not detected, using default CPU batch-size {batch_size}')
+ return batch_size
+
+ d = str(device).upper() # 'CUDA:0'
+ properties = torch.cuda.get_device_properties(device) # device properties
+ t = properties.total_memory / 1024 ** 3 # (GiB)
+ r = torch.cuda.memory_reserved(device) / 1024 ** 3 # (GiB)
+ a = torch.cuda.memory_allocated(device) / 1024 ** 3 # (GiB)
+ f = t - (r + a) # free inside reserved
+ LOGGER.info(f'{prefix}{d} ({properties.name}) {t:.2f}G total, {r:.2f}G reserved, {a:.2f}G allocated, {f:.2f}G free')
+
+ batch_sizes = [1, 2, 4, 8, 16]
+ try:
+ img = [torch.zeros(b, 3, imgsz, imgsz) for b in batch_sizes]
+ y = profile(img, model, n=3, device=device)
+ except Exception as e:
+ LOGGER.warning(f'{prefix}{e}')
+
+ y = [x[2] for x in y if x] # memory [2]
+ batch_sizes = batch_sizes[:len(y)]
+ p = np.polyfit(batch_sizes, y, deg=1) # first degree polynomial fit
+ b = int((f * fraction - p[1]) / p[0]) # y intercept (optimal batch size)
+ LOGGER.info(f'{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%)')
+ return b
diff --git a/utils/aws/__init__.py b/utils/aws/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/utils/aws/mime.sh b/utils/aws/mime.sh
new file mode 100644
index 0000000..c319a83
--- /dev/null
+++ b/utils/aws/mime.sh
@@ -0,0 +1,26 @@
+# AWS EC2 instance startup 'MIME' script https://aws.amazon.com/premiumsupport/knowledge-center/execute-user-data-ec2/
+# This script will run on every instance restart, not only on first start
+# --- DO NOT COPY ABOVE COMMENTS WHEN PASTING INTO USERDATA ---
+
+Content-Type: multipart/mixed; boundary="//"
+MIME-Version: 1.0
+
+--//
+Content-Type: text/cloud-config; charset="us-ascii"
+MIME-Version: 1.0
+Content-Transfer-Encoding: 7bit
+Content-Disposition: attachment; filename="cloud-config.txt"
+
+#cloud-config
+cloud_final_modules:
+- [scripts-user, always]
+
+--//
+Content-Type: text/x-shellscript; charset="us-ascii"
+MIME-Version: 1.0
+Content-Transfer-Encoding: 7bit
+Content-Disposition: attachment; filename="userdata.txt"
+
+#!/bin/bash
+# --- paste contents of userdata.sh here ---
+--//
diff --git a/utils/aws/resume.py b/utils/aws/resume.py
new file mode 100644
index 0000000..b21731c
--- /dev/null
+++ b/utils/aws/resume.py
@@ -0,0 +1,40 @@
+# Resume all interrupted trainings in yolov5/ dir including DDP trainings
+# Usage: $ python utils/aws/resume.py
+
+import os
+import sys
+from pathlib import Path
+
+import torch
+import yaml
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[2] # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+ sys.path.append(str(ROOT)) # add ROOT to PATH
+
+port = 0 # --master_port
+path = Path('').resolve()
+for last in path.rglob('*/**/last.pt'):
+ ckpt = torch.load(last)
+ if ckpt['optimizer'] is None:
+ continue
+
+ # Load opt.yaml
+ with open(last.parent.parent / 'opt.yaml', errors='ignore') as f:
+ opt = yaml.safe_load(f)
+
+ # Get device count
+ d = opt['device'].split(',') # devices
+ nd = len(d) # number of devices
+ ddp = nd > 1 or (nd == 0 and torch.cuda.device_count() > 1) # distributed data parallel
+
+ if ddp: # multi-GPU
+ port += 1
+ cmd = f'python -m torch.distributed.run --nproc_per_node {nd} --master_port {port} train.py --resume {last}'
+ else: # single-GPU
+ cmd = f'python train.py --resume {last}'
+
+ cmd += ' > /dev/null 2>&1 &' # redirect output to dev/null and run in daemon thread
+ print(cmd)
+ os.system(cmd)
diff --git a/utils/aws/userdata.sh b/utils/aws/userdata.sh
new file mode 100644
index 0000000..5fc1332
--- /dev/null
+++ b/utils/aws/userdata.sh
@@ -0,0 +1,27 @@
+#!/bin/bash
+# AWS EC2 instance startup script https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/user-data.html
+# This script will run only once on first instance start (for a re-start script see mime.sh)
+# /home/ubuntu (ubuntu) or /home/ec2-user (amazon-linux) is working dir
+# Use >300 GB SSD
+
+cd home/ubuntu
+if [ ! -d yolov5 ]; then
+ echo "Running first-time script." # install dependencies, download COCO, pull Docker
+ git clone https://github.com/ultralytics/yolov5 -b master && sudo chmod -R 777 yolov5
+ cd yolov5
+ bash data/scripts/get_coco.sh && echo "COCO done." &
+ sudo docker pull ultralytics/yolov5:latest && echo "Docker done." &
+ python -m pip install --upgrade pip && pip install -r requirements.txt && python detect.py && echo "Requirements done." &
+ wait && echo "All tasks done." # finish background tasks
+else
+ echo "Running re-start script." # resume interrupted runs
+ i=0
+ list=$(sudo docker ps -qa) # container list i.e. $'one\ntwo\nthree\nfour'
+ while IFS= read -r id; do
+ ((i++))
+ echo "restarting container $i: $id"
+ sudo docker start $id
+ # sudo docker exec -it $id python train.py --resume # single-GPU
+ sudo docker exec -d $id python utils/aws/resume.py # multi-scenario
+ done <<<"$list"
+fi
diff --git a/utils/callbacks.py b/utils/callbacks.py
new file mode 100644
index 0000000..13d82eb
--- /dev/null
+++ b/utils/callbacks.py
@@ -0,0 +1,77 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Callback utils
+"""
+
+
+class Callbacks:
+ """"
+ Handles all registered callbacks for YOLOv5 Hooks
+ """
+
+ def __init__(self):
+ # Define the available callbacks
+ self._callbacks = {
+ 'on_pretrain_routine_start': [],
+ 'on_pretrain_routine_end': [],
+
+ 'on_train_start': [],
+ 'on_train_epoch_start': [],
+ 'on_train_batch_start': [],
+ 'optimizer_step': [],
+ 'on_before_zero_grad': [],
+ 'on_train_batch_end': [],
+ 'on_train_epoch_end': [],
+
+ 'on_val_start': [],
+ 'on_val_batch_start': [],
+ 'on_val_image_end': [],
+ 'on_val_batch_end': [],
+ 'on_val_end': [],
+
+ 'on_fit_epoch_end': [], # fit = train + val
+ 'on_model_save': [],
+ 'on_train_end': [],
+ 'on_params_update': [],
+ 'teardown': [],
+ }
+
+ def register_action(self, hook, name='', callback=None):
+ """
+ Register a new action to a callback hook
+
+ Args:
+ hook The callback hook name to register the action to
+ name The name of the action for later reference
+ callback The callback to fire
+ """
+ assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
+ assert callable(callback), f"callback '{callback}' is not callable"
+ self._callbacks[hook].append({'name': name, 'callback': callback})
+
+ def get_registered_actions(self, hook=None):
+ """"
+ Returns all the registered actions by callback hook
+
+ Args:
+ hook The name of the hook to check, defaults to all
+ """
+ if hook:
+ return self._callbacks[hook]
+ else:
+ return self._callbacks
+
+ def run(self, hook, *args, **kwargs):
+ """
+ Loop through the registered actions and fire all callbacks
+
+ Args:
+ hook The name of the hook to check, defaults to all
+ args Arguments to receive from YOLOv5
+ kwargs Keyword Arguments to receive from YOLOv5
+ """
+
+ assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
+
+ for logger in self._callbacks[hook]:
+ logger['callback'](*args, **kwargs)
diff --git a/utils/datasets.py b/utils/datasets.py
new file mode 100644
index 0000000..fa75e04
--- /dev/null
+++ b/utils/datasets.py
@@ -0,0 +1,1110 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Dataloaders and dataset utils
+"""
+
+import glob
+import hashlib
+import json
+import os
+import random
+import shutil
+import time
+from itertools import repeat
+from multiprocessing.pool import Pool, ThreadPool
+from pathlib import Path
+from threading import Thread
+from zipfile import ZipFile
+
+import cv2
+import numpy as np
+import torch
+import torch.nn.functional as F
+import yaml
+from PIL import ExifTags, Image, ImageOps
+from torch.utils.data import DataLoader, Dataset, dataloader, distributed
+from tqdm import tqdm
+
+from utils.augmentations import Albumentations, augment_hsv, copy_paste, letterbox, mixup, random_perspective
+from utils.general import (LOGGER, NUM_THREADS, check_dataset, check_requirements, check_yaml, clean_str,
+ segments2boxes, xyn2xy, xywh2xyxy, xywhn2xyxy, xyxy2xywhn)
+from utils.torch_utils import torch_distributed_zero_first
+from utils.rboxs_utils import poly_filter, poly2rbox
+
+# Parameters
+HELP_URL = 'https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data'
+IMG_FORMATS = ['bmp', 'jpg', 'jpeg', 'png', 'tif', 'tiff', 'dng', 'webp', 'mpo'] # acceptable image suffixes
+VID_FORMATS = ['mov', 'avi', 'mp4', 'mpg', 'mpeg', 'm4v', 'wmv', 'mkv'] # acceptable video suffixes
+WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1)) # DPP
+
+# Get orientation exif tag
+for orientation in ExifTags.TAGS.keys():
+ if ExifTags.TAGS[orientation] == 'Orientation':
+ break
+
+
+def get_hash(paths):
+ # Returns a single hash value of a list of paths (files or dirs)
+ size = sum(os.path.getsize(p) for p in paths if os.path.exists(p)) # sizes
+ h = hashlib.md5(str(size).encode()) # hash sizes
+ h.update(''.join(paths).encode()) # hash paths
+ return h.hexdigest() # return hash
+
+
+def exif_size(img):
+ # Returns exif-corrected PIL size
+ s = img.size # (width, height)
+ try:
+ rotation = dict(img._getexif().items())[orientation]
+ if rotation == 6: # rotation 270
+ s = (s[1], s[0])
+ elif rotation == 8: # rotation 90
+ s = (s[1], s[0])
+ except:
+ pass
+
+ return s
+
+
+def exif_transpose(image):
+ """
+ Transpose a PIL image accordingly if it has an EXIF Orientation tag.
+ Inplace version of https://github.com/python-pillow/Pillow/blob/master/src/PIL/ImageOps.py exif_transpose()
+
+ :param image: The image to transpose.
+ :return: An image.
+ """
+ exif = image.getexif()
+ orientation = exif.get(0x0112, 1) # default 1
+ if orientation > 1:
+ method = {2: Image.FLIP_LEFT_RIGHT,
+ 3: Image.ROTATE_180,
+ 4: Image.FLIP_TOP_BOTTOM,
+ 5: Image.TRANSPOSE,
+ 6: Image.ROTATE_270,
+ 7: Image.TRANSVERSE,
+ 8: Image.ROTATE_90,
+ }.get(orientation)
+ if method is not None:
+ image = image.transpose(method)
+ del exif[0x0112]
+ image.info["exif"] = exif.tobytes()
+ return image
+
+
+def create_dataloader(path, imgsz, batch_size, stride, names, single_cls=False, hyp=None, augment=False, cache=False, pad=0.0,
+ rect=False, rank=-1, workers=8, image_weights=False, quad=False, prefix='', shuffle=False):
+ if rect and shuffle:
+ LOGGER.warning('WARNING: --rect is incompatible with DataLoader shuffle, setting shuffle=False')
+ shuffle = False
+ with torch_distributed_zero_first(rank): # init dataset *.cache only once if DDP
+ dataset = LoadImagesAndLabels(path, names, imgsz, batch_size,
+ augment=augment, # augmentation
+ hyp=hyp, # hyperparameters
+ rect=rect, # rectangular batches
+ cache_images=cache,
+ single_cls=single_cls,
+ stride=int(stride),
+ pad=pad,
+ image_weights=image_weights,
+ prefix=prefix)
+
+ batch_size = min(batch_size, len(dataset))
+ nw = min([os.cpu_count() // WORLD_SIZE, batch_size if batch_size > 1 else 0, workers]) # number of workers
+ sampler = None if rank == -1 else distributed.DistributedSampler(dataset, shuffle=shuffle)
+ loader = DataLoader if image_weights else InfiniteDataLoader # only DataLoader allows for attribute updates
+ return loader(dataset,
+ batch_size=batch_size,
+ shuffle=shuffle and sampler is None,
+ num_workers=nw,
+ sampler=sampler,
+ pin_memory=True,
+ collate_fn=LoadImagesAndLabels.collate_fn4 if quad else LoadImagesAndLabels.collate_fn), dataset
+
+
+class InfiniteDataLoader(dataloader.DataLoader):
+ """ Dataloader that reuses workers
+
+ Uses same syntax as vanilla DataLoader
+ """
+
+ def __init__(self, *args, **kwargs):
+ super().__init__(*args, **kwargs)
+ object.__setattr__(self, 'batch_sampler', _RepeatSampler(self.batch_sampler))
+ self.iterator = super().__iter__()
+
+ def __len__(self):
+ return len(self.batch_sampler.sampler)
+
+ def __iter__(self):
+ for i in range(len(self)):
+ yield next(self.iterator)
+
+
+class _RepeatSampler:
+ """ Sampler that repeats forever
+
+ Args:
+ sampler (Sampler)
+ """
+
+ def __init__(self, sampler):
+ self.sampler = sampler
+
+ def __iter__(self):
+ while True:
+ yield from iter(self.sampler)
+
+
+class LoadImages:
+ # YOLOv5 image/video dataloader, i.e. `python detect.py --source image.jpg/vid.mp4`
+ def __init__(self, path, img_size=640, stride=32, auto=True):
+ p = str(Path(path).resolve()) # os-agnostic absolute path
+ if '*' in p:
+ files = sorted(glob.glob(p, recursive=True)) # glob
+ elif os.path.isdir(p):
+ files = sorted(glob.glob(os.path.join(p, '*.*'))) # dir
+ elif os.path.isfile(p):
+ files = [p] # files
+ else:
+ raise Exception(f'ERROR: {p} does not exist')
+
+ images = [x for x in files if x.split('.')[-1].lower() in IMG_FORMATS]
+ videos = [x for x in files if x.split('.')[-1].lower() in VID_FORMATS]
+ ni, nv = len(images), len(videos)
+
+ self.img_size = img_size
+ self.stride = stride
+ self.files = images + videos
+ self.nf = ni + nv # number of files
+ self.video_flag = [False] * ni + [True] * nv
+ self.mode = 'image'
+ self.auto = auto
+ if any(videos):
+ self.new_video(videos[0]) # new video
+ else:
+ self.cap = None
+ assert self.nf > 0, f'No images or videos found in {p}. ' \
+ f'Supported formats are:\nimages: {IMG_FORMATS}\nvideos: {VID_FORMATS}'
+
+ def __iter__(self):
+ self.count = 0
+ return self
+
+ def __next__(self):
+ if self.count == self.nf:
+ raise StopIteration
+ path = self.files[self.count]
+
+ if self.video_flag[self.count]:
+ # Read video
+ self.mode = 'video'
+ ret_val, img0 = self.cap.read()
+ while not ret_val:
+ self.count += 1
+ self.cap.release()
+ if self.count == self.nf: # last video
+ raise StopIteration
+ else:
+ path = self.files[self.count]
+ self.new_video(path)
+ ret_val, img0 = self.cap.read()
+
+ self.frame += 1
+ s = f'video {self.count + 1}/{self.nf} ({self.frame}/{self.frames}) {path}: '
+
+ else:
+ # Read image
+ self.count += 1
+ img0 = cv2.imread(path) # BGR
+ assert img0 is not None, f'Image Not Found {path}'
+ s = f'image {self.count}/{self.nf} {path}: '
+
+ # Padded resize
+ img = letterbox(img0, self.img_size, stride=self.stride, auto=self.auto)[0]
+
+ # Convert
+ img = img.transpose((2, 0, 1))[::-1] # HWC to CHW, BGR to RGB
+ img = np.ascontiguousarray(img)
+
+ return path, img, img0, self.cap, s
+
+ def new_video(self, path):
+ self.frame = 0
+ self.cap = cv2.VideoCapture(path)
+ self.frames = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT))
+
+ def __len__(self):
+ return self.nf # number of files
+
+
+class LoadWebcam: # for inference
+ # YOLOv5 local webcam dataloader, i.e. `python detect.py --source 0`
+ def __init__(self, pipe='0', img_size=640, stride=32):
+ self.img_size = img_size
+ self.stride = stride
+ self.pipe = eval(pipe) if pipe.isnumeric() else pipe
+ self.cap = cv2.VideoCapture(self.pipe) # video capture object
+ self.cap.set(cv2.CAP_PROP_BUFFERSIZE, 3) # set buffer size
+
+ def __iter__(self):
+ self.count = -1
+ return self
+
+ def __next__(self):
+ self.count += 1
+ if cv2.waitKey(1) == ord('q'): # q to quit
+ self.cap.release()
+ cv2.destroyAllWindows()
+ raise StopIteration
+
+ # Read frame
+ ret_val, img0 = self.cap.read()
+ img0 = cv2.flip(img0, 1) # flip left-right
+
+ # Print
+ assert ret_val, f'Camera Error {self.pipe}'
+ img_path = 'webcam.jpg'
+ s = f'webcam {self.count}: '
+
+ # Padded resize
+ img = letterbox(img0, self.img_size, stride=self.stride)[0]
+
+ # Convert
+ img = img.transpose((2, 0, 1))[::-1] # HWC to CHW, BGR to RGB
+ img = np.ascontiguousarray(img)
+
+ return img_path, img, img0, None, s
+
+ def __len__(self):
+ return 0
+
+
+class LoadStreams:
+ # YOLOv5 streamloader, i.e. `python detect.py --source 'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP streams`
+ def __init__(self, sources='streams.txt', img_size=640, stride=32, auto=True):
+ self.mode = 'stream'
+ self.img_size = img_size
+ self.stride = stride
+
+ if os.path.isfile(sources):
+ with open(sources) as f:
+ sources = [x.strip() for x in f.read().strip().splitlines() if len(x.strip())]
+ else:
+ sources = [sources]
+
+ n = len(sources)
+ self.imgs, self.fps, self.frames, self.threads = [None] * n, [0] * n, [0] * n, [None] * n
+ self.sources = [clean_str(x) for x in sources] # clean source names for later
+ self.auto = auto
+ for i, s in enumerate(sources): # index, source
+ # Start thread to read frames from video stream
+ st = f'{i + 1}/{n}: {s}... '
+ if 'youtube.com/' in s or 'youtu.be/' in s: # if source is YouTube video
+ check_requirements(('pafy', 'youtube_dl'))
+ import pafy
+ s = pafy.new(s).getbest(preftype="mp4").url # YouTube URL
+ s = eval(s) if s.isnumeric() else s # i.e. s = '0' local webcam
+ cap = cv2.VideoCapture(s)
+ assert cap.isOpened(), f'{st}Failed to open {s}'
+ w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+ h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
+ self.fps[i] = max(cap.get(cv2.CAP_PROP_FPS) % 100, 0) or 30.0 # 30 FPS fallback
+ self.frames[i] = max(int(cap.get(cv2.CAP_PROP_FRAME_COUNT)), 0) or float('inf') # infinite stream fallback
+
+ _, self.imgs[i] = cap.read() # guarantee first frame
+ self.threads[i] = Thread(target=self.update, args=([i, cap, s]), daemon=True)
+ LOGGER.info(f"{st} Success ({self.frames[i]} frames {w}x{h} at {self.fps[i]:.2f} FPS)")
+ self.threads[i].start()
+ LOGGER.info('') # newline
+
+ # check for common shapes
+ s = np.stack([letterbox(x, self.img_size, stride=self.stride, auto=self.auto)[0].shape for x in self.imgs])
+ self.rect = np.unique(s, axis=0).shape[0] == 1 # rect inference if all shapes equal
+ if not self.rect:
+ LOGGER.warning('WARNING: Stream shapes differ. For optimal performance supply similarly-shaped streams.')
+
+ def update(self, i, cap, stream):
+ # Read stream `i` frames in daemon thread
+ n, f, read = 0, self.frames[i], 1 # frame number, frame array, inference every 'read' frame
+ while cap.isOpened() and n < f:
+ n += 1
+ # _, self.imgs[index] = cap.read()
+ cap.grab()
+ if n % read == 0:
+ success, im = cap.retrieve()
+ if success:
+ self.imgs[i] = im
+ else:
+ LOGGER.warning('WARNING: Video stream unresponsive, please check your IP camera connection.')
+ self.imgs[i] = np.zeros_like(self.imgs[i])
+ cap.open(stream) # re-open stream if signal was lost
+ time.sleep(1 / self.fps[i]) # wait time
+
+ def __iter__(self):
+ self.count = -1
+ return self
+
+ def __next__(self):
+ self.count += 1
+ if not all(x.is_alive() for x in self.threads) or cv2.waitKey(1) == ord('q'): # q to quit
+ cv2.destroyAllWindows()
+ raise StopIteration
+
+ # Letterbox
+ img0 = self.imgs.copy()
+ img = [letterbox(x, self.img_size, stride=self.stride, auto=self.rect and self.auto)[0] for x in img0]
+
+ # Stack
+ img = np.stack(img, 0)
+
+ # Convert
+ img = img[..., ::-1].transpose((0, 3, 1, 2)) # BGR to RGB, BHWC to BCHW
+ img = np.ascontiguousarray(img)
+
+ return self.sources, img, img0, None, ''
+
+ def __len__(self):
+ return len(self.sources) # 1E12 frames = 32 streams at 30 FPS for 30 years
+
+
+def img2label_paths(img_paths):
+ # Define label paths as a function of image paths
+ sa, sb = os.sep + 'images' + os.sep, os.sep + 'labelTxt' + os.sep # /images/, /labels/ substrings
+ return [sb.join(x.rsplit(sa, 1)).rsplit('.', 1)[0] + '.txt' for x in img_paths]
+
+
+class LoadImagesAndLabels(Dataset):
+ # YOLOv5 train_loader/val_loader, loads images and labels for training and validation
+ cache_version = 0.6 # dataset labels *.cache version
+
+ def __init__(self, path, cls_names, img_size=640, batch_size=16, augment=False, hyp=None, rect=False, image_weights=False,
+ cache_images=False, single_cls=False, stride=32, pad=0.0, prefix=''):
+ """
+ Returns:
+ Dataset.labels (list): n_imgs * array(num_gt_perimg, [cls_id, poly])
+ Dataset.shapes (array): (n_imgs, [ori_img_width, ori_img_height])
+
+ Dataset.batch_shapes (array): (n_batches, [h_rect, w_rect])
+ """
+ self.img_size = img_size
+ self.augment = augment
+ self.hyp = hyp
+ self.image_weights = image_weights
+ self.rect = False if image_weights else rect
+ self.mosaic = self.augment and not self.rect # load 4 images at a time into a mosaic (only during training)
+ self.mosaic_border = [-img_size // 2, -img_size // 2]
+ self.stride = stride
+ self.path = path
+ self.albumentations = Albumentations() if augment else None
+ self.cls_names = cls_names
+
+ try:
+ f = [] # image files
+ for p in path if isinstance(path, list) else [path]:
+ p = Path(p) # os-agnostic
+ if p.is_dir(): # dir
+ f += glob.glob(str(p / '**' / '*.*'), recursive=True)
+ # f = list(p.rglob('*.*')) # pathlib
+ elif p.is_file(): # file
+ with open(p) as t:
+ t = t.read().strip().splitlines()
+ parent = str(p.parent) + os.sep
+ f += [x.replace('./', parent) if x.startswith('./') else x for x in t] # local to global path
+ # f += [p.parent / x.lstrip(os.sep) for x in t] # local to global path (pathlib)
+ else:
+ raise Exception(f'{prefix}{p} does not exist')
+ self.img_files = sorted(x.replace('/', os.sep) for x in f if x.split('.')[-1].lower() in IMG_FORMATS)
+ # self.img_files = sorted([x for x in f if x.suffix[1:].lower() in IMG_FORMATS]) # pathlib
+ assert self.img_files, f'{prefix}No images found'
+ except Exception as e:
+ raise Exception(f'{prefix}Error loading data from {path}: {e}\nSee {HELP_URL}')
+
+ # Check cache
+ self.label_files = img2label_paths(self.img_files) # labels
+ cache_path = (p if p.is_file() else Path(self.label_files[0]).parent).with_suffix('.cache')
+ try:
+ cache, exists = np.load(cache_path, allow_pickle=True).item(), True # load dict
+ assert cache['version'] == self.cache_version # same version
+ assert cache['hash'] == get_hash(self.label_files + self.img_files) # same hash
+ except:
+ cache, exists = self.cache_labels(cache_path, prefix), False # cache
+
+ # Display cache
+ nf, nm, ne, nc, n = cache.pop('results') # found, missing, empty, corrupted, total
+ if exists:
+ d = f"Scanning '{cache_path}' images and labels... {nf} found, {nm} missing, {ne} empty, {nc} corrupted"
+ tqdm(None, desc=prefix + d, total=n, initial=n) # display cache results
+ if cache['msgs']:
+ LOGGER.info('\n'.join(cache['msgs'])) # display warnings
+ assert nf > 0 or not augment, f'{prefix}No labels in {cache_path}. Can not train without labels. See {HELP_URL}'
+
+ # Read cache
+ [cache.pop(k) for k in ('hash', 'version', 'msgs')] # remove items
+ labels, shapes, self.segments = zip(*cache.values())
+ self.labels = list(labels) # labels(list[array]): n_imgs * array(num_gt_perimg, [cls_id, poly])
+ self.shapes = np.array(shapes, dtype=np.float64) # img_ori shape
+ self.img_files = list(cache.keys()) # update
+ self.label_files = img2label_paths(cache.keys()) # update
+ n = len(shapes) # number of images
+ bi = np.floor(np.arange(n) / batch_size).astype(np.int) # batch index
+ nb = bi[-1] + 1 # number of batches
+ self.batch = bi # batch index of image
+ self.n = n
+ self.indices = range(n)
+
+ # Update labels
+ include_class = [] # filter labels to include only these classes (optional)
+ include_class_array = np.array(include_class).reshape(1, -1)
+ for i, (label, segment) in enumerate(zip(self.labels, self.segments)):
+ if include_class:
+ j = (label[:, 0:1] == include_class_array).any(1)
+ self.labels[i] = label[j]
+ if segment:
+ self.segments[i] = segment[j]
+ if single_cls: # single-class training, merge all classes into 0
+ self.labels[i][:, 0] = 0
+ if segment:
+ self.segments[i][:, 0] = 0
+
+ # Rectangular Training
+ if self.rect:
+ # Sort by aspect ratio
+ s = self.shapes # wh
+ ar = s[:, 1] / s[:, 0] # aspect ratio
+ irect = ar.argsort()
+ self.img_files = [self.img_files[i] for i in irect]
+ self.label_files = [self.label_files[i] for i in irect]
+ self.labels = [self.labels[i] for i in irect]
+ self.shapes = s[irect] # wh
+ ar = ar[irect]
+
+ # Set training image shapes
+ shapes = [[1, 1]] * nb
+ for i in range(nb):
+ ari = ar[bi == i]
+ mini, maxi = ari.min(), ari.max()
+ if maxi < 1: # batch图像高宽比均小于1时, shape=[h/w, 1] = [h_ratio, w_ratio]
+ shapes[i] = [maxi, 1]
+ elif mini > 1: # batch图像高宽比均大于1时, shape=[1, w/h] = [h_ratio, w_ratio]
+ shapes[i] = [1, 1 / mini]
+
+ self.batch_shapes = np.ceil(np.array(shapes) * img_size / stride + pad).astype(np.int) * stride # (nb, [h_rect, w_rect])
+
+ # Cache images into memory for faster training (WARNING: large datasets may exceed system RAM)
+ self.imgs, self.img_npy = [None] * n, [None] * n
+ if cache_images:
+ if cache_images == 'disk':
+ self.im_cache_dir = Path(Path(self.img_files[0]).parent.as_posix() + '_npy')
+ self.img_npy = [self.im_cache_dir / Path(f).with_suffix('.npy').name for f in self.img_files]
+ self.im_cache_dir.mkdir(parents=True, exist_ok=True)
+ gb = 0 # Gigabytes of cached images
+ self.img_hw0, self.img_hw = [None] * n, [None] * n
+ results = ThreadPool(NUM_THREADS).imap(lambda x: load_image_label(*x), zip(repeat(self), range(n)))
+ pbar = tqdm(enumerate(results), total=n)
+ for i, x in pbar:
+ if cache_images == 'disk':
+ if not self.img_npy[i].exists():
+ np.save(self.img_npy[i].as_posix(), x[0])
+ gb += self.img_npy[i].stat().st_size
+ else:
+ self.imgs[i], self.img_hw0[i], self.img_hw[i], self.labels[i] = x # im, hw_orig, hw_resized, label_resized = load_image_label(self, i)
+ gb += self.imgs[i].nbytes
+ pbar.desc = f'{prefix}Caching images ({gb / 1E9:.1f}GB {cache_images})'
+ pbar.close()
+
+ def cache_labels(self, path=Path('./labels.cache'), prefix=''):
+ # Cache dataset labels, check images and read shapes
+ x = {} # dict
+ nm, nf, ne, nc, msgs = 0, 0, 0, 0, [] # number missing, found, empty, corrupt, messages
+ desc = f"{prefix}Scanning '{path.parent / path.stem}' images and labels..."
+ with Pool(NUM_THREADS) as pool:
+ pbar = tqdm(pool.imap(verify_image_label, zip(self.img_files, self.label_files, repeat(prefix), repeat(self.cls_names))),
+ desc=desc, total=len(self.img_files))
+ for im_file, l, shape, segments, nm_f, nf_f, ne_f, nc_f, msg in pbar:
+ nm += nm_f
+ nf += nf_f
+ ne += ne_f
+ nc += nc_f
+ if im_file:
+ x[im_file] = [l, shape, segments]
+ if msg:
+ msgs.append(msg)
+ pbar.desc = f"{desc}{nf} found, {nm} missing, {ne} empty, {nc} corrupted"
+
+ pbar.close()
+ if msgs:
+ LOGGER.info('\n'.join(msgs))
+ if nf == 0:
+ LOGGER.warning(f'{prefix}WARNING: No labels found in {path}. See {HELP_URL}')
+ x['hash'] = get_hash(self.label_files + self.img_files)
+ x['results'] = nf, nm, ne, nc, len(self.img_files)
+ x['msgs'] = msgs # warnings
+ x['version'] = self.cache_version # cache version
+ try:
+ np.save(path, x) # save cache for next time
+ path.with_suffix('.cache.npy').rename(path) # remove .npy suffix
+ LOGGER.info(f'{prefix}New cache created: {path}')
+ except Exception as e:
+ LOGGER.warning(f'{prefix}WARNING: Cache directory {path.parent} is not writeable: {e}') # not writeable
+ return x
+
+ def __len__(self):
+ return len(self.img_files)
+
+ # def __iter__(self):
+ # self.count = -1
+ # print('ran dataset iter')
+ # #self.shuffled_vector = np.random.permutation(self.nF) if self.augment else np.arange(self.nF)
+ # return self
+
+ def __getitem__(self, index):
+ '''
+ Augment the [clsid poly] labels and trans label format to rbox.
+ Returns:
+ img (tensor): (3, height, width), RGB
+ labels_out (tensor): (n, [None clsid cx cy l s theta gaussian_θ_labels]) θ∈[-pi/2, pi/2)
+ img_file (str): img_dir
+ shapes : None or [(h_raw, w_raw), (hw_ratios, wh_paddings)], for COCO mAP rescaling
+ '''
+ index = self.indices[index] # linear, shuffled, or image_weights
+
+ hyp = self.hyp
+ mosaic = self.mosaic and random.random() < hyp['mosaic']
+ if mosaic:
+ # Load mosaic
+ img, labels = load_mosaic(self, index)
+ shapes = None
+
+ # MixUp augmentation
+ if random.random() < hyp['mixup']:
+ img, labels = mixup(img, labels, *load_mosaic(self, random.randint(0, self.n - 1)))
+
+ else:
+ # Load image and label
+ img, (h0, w0), (h, w), img_label = load_image_label(self, index)
+
+ # Letterbox
+ shape = self.batch_shapes[self.batch[index]] if self.rect else self.img_size # final letterboxed shape [h_rect, w_rect]
+ img, ratio, pad = letterbox(img, shape, auto=False, scaleup=self.augment) # ratio[w_ratio, h_ratio], pad[w_padding, h_padding]
+ shapes = (h0, w0), ((h / h0, w / w0), pad) # for COCO mAP rescaling [(h_raw, w_raw), (hw_ratios, wh_paddings)]
+
+ labels = img_label.copy() # labels (array): (num_gt_perimg, [cls_id, poly])
+ if labels.size:
+ # labels[:, 1:] = xywhn2xyxy(labels[:, 1:], ratio[0] * w, ratio[1] * h, padw=pad[0], padh=pad[1])
+ labels[:, [1, 3, 5, 7]] = img_label[:, [1, 3, 5, 7]] * ratio[0] + pad[0]
+ labels[:, [2, 4, 6, 8]] = img_label[:, [2, 4, 6, 8]] * ratio[1] + pad[1]
+
+ if self.augment:
+ img, labels = random_perspective(img, labels,
+ degrees=hyp['degrees'],
+ translate=hyp['translate'],
+ scale=hyp['scale'],
+ shear=hyp['shear'],
+ perspective=hyp['perspective'])
+
+ nl = len(labels) # number of labels
+ # if nl:
+ # labels[:, 1:5] = xyxy2xywhn(labels[:, 1:5], w=img.shape[1], h=img.shape[0], clip=True, eps=1E-3)
+
+
+ if self.augment:
+ # Albumentations
+ # img, labels = self.albumentations(img, labels)
+ # nl = len(labels) # update after albumentations
+
+ # HSV color-space
+ augment_hsv(img, hgain=hyp['hsv_h'], sgain=hyp['hsv_s'], vgain=hyp['hsv_v'])
+
+ img_h, img_w = img.shape[0], img.shape[1]
+ # Flip up-down
+ if random.random() < hyp['flipud']:
+ img = np.flipud(img)
+ if nl:
+ # labels[:, 2] = 1 - labels[:, 2]
+ labels[:, 2::2] = img_h - labels[:, 2::2] - 1
+
+ # Flip left-right
+ if random.random() < hyp['fliplr']:
+ img = np.fliplr(img)
+ if nl:
+ # labels[:, 1] = 1 - labels[:, 1]
+ labels[:, 1::2] = img_w - labels[:, 1::2] - 1
+
+ # Cutouts
+ # labels = cutout(img, labels, p=0.5)
+ # nl = len(labels) # update after cutout
+ if nl:
+ # *[clsid poly] to *[clsid cx cy l s theta gaussian_θ_labels] θ∈[-pi/2, pi/2) non-normalized
+ rboxes, csl_labels = poly2rbox(polys=labels[:, 1:],
+ num_cls_thata=hyp['cls_theta'] if hyp else 180,
+ radius=hyp['csl_radius'] if hyp else 6.0,
+ use_pi=True, use_gaussian=True)
+ labels_obb = np.concatenate((labels[:, :1], rboxes, csl_labels), axis=1)
+ labels_mask = (rboxes[:, 0] >= 0) & (rboxes[:, 0] < img.shape[1]) \
+ & (rboxes[:, 1] >= 0) & (rboxes[:, 0] < img.shape[0]) \
+ & (rboxes[:, 2] > 5) | (rboxes[:, 3] > 5)
+ labels_obb = labels_obb[labels_mask]
+ nl = len(labels_obb) # update after filter
+
+ if hyp:
+ c_num = 7 + hyp['cls_theta'] # [index_of_batch clsid cx cy l s theta gaussian_θ_labels]
+ else:
+ c_num = 187
+
+ # labels_out = torch.zeros((nl, 6))
+ labels_out = torch.zeros((nl, c_num))
+ if nl:
+ # labels_out[:, 1:] = torch.from_numpy(labels)
+ labels_out[:, 1:] = torch.from_numpy(labels_obb)
+
+ # Convert
+ img = img.transpose((2, 0, 1))[::-1] # HWC to CHW, BGR to RGB
+ img = np.ascontiguousarray(img)
+
+ return torch.from_numpy(img), labels_out, self.img_files[index], shapes
+
+ @staticmethod
+ def collate_fn(batch):
+ img, label, path, shapes = zip(*batch) # transposed; (tupe(b*tensor))
+ for i, l in enumerate(label):
+ l[:, 0] = i # add target image index for build_targets()
+ return torch.stack(img, 0), torch.cat(label, 0), path, shapes
+
+ @staticmethod
+ def collate_fn4(batch):
+ img, label, path, shapes = zip(*batch) # transposed
+ n = len(shapes) // 4
+ img4, label4, path4, shapes4 = [], [], path[:n], shapes[:n]
+
+ ho = torch.tensor([[0.0, 0, 0, 1, 0, 0]])
+ wo = torch.tensor([[0.0, 0, 1, 0, 0, 0]])
+ s = torch.tensor([[1, 1, 0.5, 0.5, 0.5, 0.5]]) # scale
+ for i in range(n): # zidane torch.zeros(16,3,720,1280) # BCHW
+ i *= 4
+ if random.random() < 0.5:
+ im = F.interpolate(img[i].unsqueeze(0).float(), scale_factor=2.0, mode='bilinear', align_corners=False)[
+ 0].type(img[i].type())
+ l = label[i]
+ else:
+ im = torch.cat((torch.cat((img[i], img[i + 1]), 1), torch.cat((img[i + 2], img[i + 3]), 1)), 2)
+ l = torch.cat((label[i], label[i + 1] + ho, label[i + 2] + wo, label[i + 3] + ho + wo), 0) * s
+ img4.append(im)
+ label4.append(l)
+
+ for i, l in enumerate(label4):
+ l[:, 0] = i # add target image index for build_targets()
+
+ return torch.stack(img4, 0), torch.cat(label4, 0), path4, shapes4
+
+
+# Ancillary functions --------------------------------------------------------------------------------------------------
+def load_image_label(self, i):
+ # loads 1 image from dataset index 'i', returns im, original hw, resized hw
+ im = self.imgs[i]
+ label = self.labels[i].copy() # labels (array): (num_gt_perimg, [cls_id, poly])
+ if im is None: # not cached in ram
+ npy = self.img_npy[i]
+ if npy and npy.exists(): # load npy
+ im = np.load(npy)
+ else: # read image
+ path = self.img_files[i]
+ im = cv2.imread(path) # BGR
+ assert im is not None, f'Image Not Found {path}'
+ h0, w0 = im.shape[:2] # orig hw
+ r = self.img_size / max(h0, w0) # ratio
+ if r != 1: # if sizes are not equal
+ im = cv2.resize(im, (int(w0 * r), int(h0 * r)),
+ interpolation=cv2.INTER_AREA if r < 1 and not self.augment else cv2.INTER_LINEAR)
+ label[:, 1:] *= r
+ return im, (h0, w0), im.shape[:2], label # im, hw_original, hw_resized, resized_label
+ else:
+ return self.imgs[i], self.img_hw0[i], self.img_hw[i], self.labels[i] # im, hw_original, hw_resized, resized_label
+
+
+def load_mosaic(self, index):
+ # YOLOv5 4-mosaic loader. Loads 1 image + 3 random images into a 4-image mosaic
+ labels4, segments4 = [], []
+ s = self.img_size
+ yc, xc = (int(random.uniform(-x, 2 * s + x)) for x in self.mosaic_border) # mosaic center x, y
+ indices = [index] + random.choices(self.indices, k=3) # 3 additional image indices
+ random.shuffle(indices)
+ for i, index in enumerate(indices):
+ # Load image
+ img, _, (h, w), img_label = load_image_label(self, index)
+
+ # place img in img4
+ if i == 0: # top left
+ img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles
+ x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc # xmin, ymin, xmax, ymax (large image)
+ x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h # xmin, ymin, xmax, ymax (small image)
+ elif i == 1: # top right
+ x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
+ x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
+ elif i == 2: # bottom left
+ x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
+ x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h)
+ elif i == 3: # bottom right
+ x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
+ x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
+
+ img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax]
+ padw = x1a - x1b
+ padh = y1a - y1b
+
+ # Labels
+ labels, segments = img_label.copy(), self.segments[index].copy() # labels (array): (num_gt_perimg, [cls_id, poly])
+ if labels.size:
+ # labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padw, padh) # normalized xywh to pixel xyxy format
+ labels[:, [1, 3, 5, 7]] = img_label[:, [1, 3, 5, 7]] + padw
+ labels[:, [2, 4, 6, 8]] = img_label[:, [2, 4, 6, 8]] + padh
+ segments = [xyn2xy(x, w, h, padw, padh) for x in segments]
+ labels4.append(labels)
+ segments4.extend(segments)
+
+ # Concat/clip labels
+ labels4 = np.concatenate(labels4, 0)
+ # for x in (labels4[:, 1:], *segments4):
+ for x in (segments4):
+ np.clip(x, 0, 2 * s, out=x) # clip when using random_perspective()
+ h_filter = 2 * s
+ w_filter = 2 * s
+ labels_mask = poly_filter(polys=labels4[:, 1:].copy(), h=h_filter, w=w_filter)
+ labels4 = labels4[labels_mask]
+ # img4, labels4 = replicate(img4, labels4) # replicate
+
+ # Augment
+ img4, labels4, segments4 = copy_paste(img4, labels4, segments4, p=self.hyp['copy_paste'])
+ img4, labels4 = random_perspective(img4, labels4, segments4,
+ degrees=self.hyp['degrees'],
+ translate=self.hyp['translate'],
+ scale=self.hyp['scale'],
+ shear=self.hyp['shear'],
+ perspective=self.hyp['perspective'],
+ border=self.mosaic_border) # border to remove
+
+ return img4, labels4
+
+
+def load_mosaic9(self, index):
+ # YOLOv5 9-mosaic loader. Loads 1 image + 8 random images into a 9-image mosaic
+ labels9, segments9 = [], []
+ s = self.img_size
+ indices = [index] + random.choices(self.indices, k=8) # 8 additional image indices
+ random.shuffle(indices)
+ for i, index in enumerate(indices):
+ # Load image
+ img, _, (h, w), img_label = load_image_label(self, index)
+
+ # place img in img9
+ if i == 0: # center
+ img9 = np.full((s * 3, s * 3, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles
+ h0, w0 = h, w
+ c = s, s, s + w, s + h # xmin, ymin, xmax, ymax (base) coordinates
+ elif i == 1: # top
+ c = s, s - h, s + w, s
+ elif i == 2: # top right
+ c = s + wp, s - h, s + wp + w, s
+ elif i == 3: # right
+ c = s + w0, s, s + w0 + w, s + h
+ elif i == 4: # bottom right
+ c = s + w0, s + hp, s + w0 + w, s + hp + h
+ elif i == 5: # bottom
+ c = s + w0 - w, s + h0, s + w0, s + h0 + h
+ elif i == 6: # bottom left
+ c = s + w0 - wp - w, s + h0, s + w0 - wp, s + h0 + h
+ elif i == 7: # left
+ c = s - w, s + h0 - h, s, s + h0
+ elif i == 8: # top left
+ c = s - w, s + h0 - hp - h, s, s + h0 - hp
+
+ padx, pady = c[:2]
+ x1, y1, x2, y2 = (max(x, 0) for x in c) # allocate coords
+
+ # Labels
+ labels, segments = img_label.copy(), self.segments[index].copy() # labels (array): (num_gt_perimg, [cls_id, poly])
+ if labels.size:
+ # labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padx, pady) # normalized xywh to pixel xyxy format
+ segments = [xyn2xy(x, w, h, padx, pady) for x in segments]
+ labels_ = labels.clone() if isinstance(labels, torch.Tensor) else np.copy(labels)
+ labels_[:, [1, 3, 5, 7]] = labels[:, [1, 3, 5, 7]] + padx
+ labels_[:, [2, 4, 6, 8]] = labels[:, [2, 4, 6, 8]] + pady
+ labels = labels_
+
+ labels9.append(labels)
+ segments9.extend(segments)
+
+ # Image
+ img9[y1:y2, x1:x2] = img[y1 - pady:, x1 - padx:] # img9[ymin:ymax, xmin:xmax]
+ hp, wp = h, w # height, width previous
+
+ # Offset
+ yc, xc = (int(random.uniform(0, s)) for _ in self.mosaic_border) # mosaic center x, y
+ img9 = img9[yc:yc + 2 * s, xc:xc + 2 * s]
+
+ # Concat/clip labels
+ labels9 = np.concatenate(labels9, 0)
+ # labels9[:, [1, 3]] -= xc
+ # labels9[:, [2, 4]] -= yc
+ labels9[:, [1, 3, 5, 7]] -= xc
+ labels9[:, [2, 4, 6, 8]] -= yc
+
+ c = np.array([xc, yc]) # centers
+ segments9 = [x - c for x in segments9]
+
+ # for x in (labels9[:, 1:], *segments9):
+ for x in (segments9):
+ np.clip(x, 0, 2 * s, out=x) # clip when using random_perspective()
+ h_filter = 2 * s
+ w_filter = 2 * s
+ labels_mask = poly_filter(polys=labels9[:, 1:].copy(), h=h_filter, w=w_filter)
+ labels9 = labels9[labels_mask]
+ # img9, labels9 = replicate(img9, labels9) # replicate
+
+ # Augment
+ img9, labels9 = random_perspective(img9, labels9, segments9,
+ degrees=self.hyp['degrees'],
+ translate=self.hyp['translate'],
+ scale=self.hyp['scale'],
+ shear=self.hyp['shear'],
+ perspective=self.hyp['perspective'],
+ border=self.mosaic_border) # border to remove
+
+ return img9, labels9
+
+
+def create_folder(path='./new'):
+ # Create folder
+ if os.path.exists(path):
+ shutil.rmtree(path) # delete output folder
+ os.makedirs(path) # make new output folder
+
+
+def flatten_recursive(path='../datasets/coco128'):
+ # Flatten a recursive directory by bringing all files to top level
+ new_path = Path(path + '_flat')
+ create_folder(new_path)
+ for file in tqdm(glob.glob(str(Path(path)) + '/**/*.*', recursive=True)):
+ shutil.copyfile(file, new_path / Path(file).name)
+
+
+def extract_boxes(path='../datasets/coco128'): # from utils.datasets import *; extract_boxes()
+ # Convert detection dataset into classification dataset, with one directory per class
+ path = Path(path) # images dir
+ shutil.rmtree(path / 'classifier') if (path / 'classifier').is_dir() else None # remove existing
+ files = list(path.rglob('*.*'))
+ n = len(files) # number of files
+ for im_file in tqdm(files, total=n):
+ if im_file.suffix[1:] in IMG_FORMATS:
+ # image
+ im = cv2.imread(str(im_file))[..., ::-1] # BGR to RGB
+ h, w = im.shape[:2]
+
+ # labels
+ lb_file = Path(img2label_paths([str(im_file)])[0])
+ if Path(lb_file).exists():
+ with open(lb_file) as f:
+ lb = np.array([x.split() for x in f.read().strip().splitlines()], dtype=np.float32) # labels
+
+ for j, x in enumerate(lb):
+ c = int(x[0]) # class
+ f = (path / 'classifier') / f'{c}' / f'{path.stem}_{im_file.stem}_{j}.jpg' # new filename
+ if not f.parent.is_dir():
+ f.parent.mkdir(parents=True)
+
+ b = x[1:] * [w, h, w, h] # box
+ # b[2:] = b[2:].max() # rectangle to square
+ b[2:] = b[2:] * 1.2 + 3 # pad
+ b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(np.int)
+
+ b[[0, 2]] = np.clip(b[[0, 2]], 0, w) # clip boxes outside of image
+ b[[1, 3]] = np.clip(b[[1, 3]], 0, h)
+ assert cv2.imwrite(str(f), im[b[1]:b[3], b[0]:b[2]]), f'box failure in {f}'
+
+
+def autosplit(path='../datasets/coco128/images', weights=(0.9, 0.1, 0.0), annotated_only=False):
+ """ Autosplit a dataset into train/val/test splits and save path/autosplit_*.txt files
+ Usage: from utils.datasets import *; autosplit()
+ Arguments
+ path: Path to images directory
+ weights: Train, val, test weights (list, tuple)
+ annotated_only: Only use images with an annotated txt file
+ """
+ path = Path(path) # images dir
+ files = sorted(x for x in path.rglob('*.*') if x.suffix[1:].lower() in IMG_FORMATS) # image files only
+ n = len(files) # number of files
+ random.seed(0) # for reproducibility
+ indices = random.choices([0, 1, 2], weights=weights, k=n) # assign each image to a split
+
+ txt = ['autosplit_train.txt', 'autosplit_val.txt', 'autosplit_test.txt'] # 3 txt files
+ [(path.parent / x).unlink(missing_ok=True) for x in txt] # remove existing
+
+ print(f'Autosplitting images from {path}' + ', using *.txt labeled images only' * annotated_only)
+ for i, img in tqdm(zip(indices, files), total=n):
+ if not annotated_only or Path(img2label_paths([str(img)])[0]).exists(): # check label
+ with open(path.parent / txt[i], 'a') as f:
+ f.write('./' + img.relative_to(path.parent).as_posix() + '\n') # add image to txt file
+
+
+def verify_image_label(args):
+ # Verify one image-label pair
+ im_file, lb_file, prefix, cls_name_list = args
+ nm, nf, ne, nc, msg, segments = 0, 0, 0, 0, '', [] # number (missing, found, empty, corrupt), message, segments
+ try:
+ # verify images
+ im = Image.open(im_file)
+ im.verify() # PIL verify
+ shape = exif_size(im) # image size
+ assert (shape[0] > 9) & (shape[1] > 9), f'image size {shape} <10 pixels'
+ assert im.format.lower() in IMG_FORMATS, f'invalid image format {im.format}'
+ if im.format.lower() in ('jpg', 'jpeg'):
+ with open(im_file, 'rb') as f:
+ f.seek(-2, 2)
+ if f.read() != b'\xff\xd9': # corrupt JPEG
+ ImageOps.exif_transpose(Image.open(im_file)).save(im_file, 'JPEG', subsampling=0, quality=100)
+ msg = f'{prefix}WARNING: {im_file}: corrupt JPEG restored and saved'
+
+ # verify labels
+ if os.path.isfile(lb_file):
+ nf = 1 # label found
+ with open(lb_file) as f:
+ labels = [x.split() for x in f.read().strip().splitlines() if len(x)]
+
+ # Yolov5-obb does not support segment labels yet
+ # if any([len(x) > 8 for x in l]): # is segment
+ # classes = np.array([x[0] for x in l], dtype=np.float32)
+ # segments = [np.array(x[1:], dtype=np.float32).reshape(-1, 2) for x in l] # (cls, xy1...)
+ # l = np.concatenate((classes.reshape(-1, 1), segments2boxes(segments)), 1) # (cls, xywh)
+ l_ = []
+ for label in labels:
+ if label[-1] == "2": # diffcult
+ continue
+ cls_id = cls_name_list.index(label[8])
+ l_.append(np.concatenate((cls_id, label[:8]), axis=None))
+ l = np.array(l_, dtype=np.float32)
+ nl = len(l)
+ if nl:
+ assert len(label) == 10, f'Yolov5-OBB labels require 10 columns, which same as DOTA Dataset, {len(label)} columns detected'
+ assert (l >= 0).all(), f'negative label values {l[l < 0]}, please check your dota format labels'
+ #assert (l[:, 1:] <= 1).all(), f'non-normalized or out of bounds coordinates {l[:, 1:][l[:, 1:] > 1]}'
+ _, i = np.unique(l, axis=0, return_index=True)
+ if len(i) < nl: # duplicate row check
+ l = l[i] # remove duplicates
+ if segments:
+ segments = segments[i]
+ msg = f'{prefix}WARNING: {im_file}: {nl - len(i)} duplicate labels removed'
+ else:
+ ne = 1 # label empty
+ # l = np.zeros((0, 5), dtype=np.float32)
+ l = np.zeros((0, 9), dtype=np.float32)
+ else:
+ nm = 1 # label missing
+ # l = np.zeros((0, 5), dtype=np.float32)
+ l = np.zeros((0, 9), dtype=np.float32)
+ return im_file, l, shape, segments, nm, nf, ne, nc, msg
+ except Exception as e:
+ nc = 1
+ msg = f'{prefix}WARNING: {im_file}: ignoring corrupt image/label: {e}'
+ return [None, None, None, None, nm, nf, ne, nc, msg]
+
+
+def dataset_stats(path='coco128.yaml', autodownload=False, verbose=False, profile=False, hub=False):
+ """ Return dataset statistics dictionary with images and instances counts per split per class
+ To run in parent directory: export PYTHONPATH="$PWD/yolov5"
+ Usage1: from utils.datasets import *; dataset_stats('coco128.yaml', autodownload=True)
+ Usage2: from utils.datasets import *; dataset_stats('../datasets/coco128_with_yaml.zip')
+ Arguments
+ path: Path to data.yaml or data.zip (with data.yaml inside data.zip)
+ autodownload: Attempt to download dataset if not found locally
+ verbose: Print stats dictionary
+ """
+
+ def round_labels(labels):
+ # Update labels to integer class and 6 decimal place floats
+ return [[int(c), *(round(x, 4) for x in points)] for c, *points in labels]
+
+ def unzip(path):
+ # Unzip data.zip TODO: CONSTRAINT: path/to/abc.zip MUST unzip to 'path/to/abc/'
+ if str(path).endswith('.zip'): # path is data.zip
+ assert Path(path).is_file(), f'Error unzipping {path}, file not found'
+ ZipFile(path).extractall(path=path.parent) # unzip
+ dir = path.with_suffix('') # dataset directory == zip name
+ return True, str(dir), next(dir.rglob('*.yaml')) # zipped, data_dir, yaml_path
+ else: # path is data.yaml
+ return False, None, path
+
+ def hub_ops(f, max_dim=1920):
+ # HUB ops for 1 image 'f': resize and save at reduced quality in /dataset-hub for web/app viewing
+ f_new = im_dir / Path(f).name # dataset-hub image filename
+ try: # use PIL
+ im = Image.open(f)
+ r = max_dim / max(im.height, im.width) # ratio
+ if r < 1.0: # image too large
+ im = im.resize((int(im.width * r), int(im.height * r)))
+ im.save(f_new, 'JPEG', quality=75, optimize=True) # save
+ except Exception as e: # use OpenCV
+ print(f'WARNING: HUB ops PIL failure {f}: {e}')
+ im = cv2.imread(f)
+ im_height, im_width = im.shape[:2]
+ r = max_dim / max(im_height, im_width) # ratio
+ if r < 1.0: # image too large
+ im = cv2.resize(im, (int(im_width * r), int(im_height * r)), interpolation=cv2.INTER_AREA)
+ cv2.imwrite(str(f_new), im)
+
+ zipped, data_dir, yaml_path = unzip(Path(path))
+ with open(check_yaml(yaml_path), errors='ignore') as f:
+ data = yaml.safe_load(f) # data dict
+ if zipped:
+ data['path'] = data_dir # TODO: should this be dir.resolve()?
+ check_dataset(data, autodownload) # download dataset if missing
+ hub_dir = Path(data['path'] + ('-hub' if hub else ''))
+ stats = {'nc': data['nc'], 'names': data['names']} # statistics dictionary
+ for split in 'train', 'val', 'test':
+ if data.get(split) is None:
+ stats[split] = None # i.e. no test set
+ continue
+ x = []
+ dataset = LoadImagesAndLabels(data[split]) # load dataset
+ for label in tqdm(dataset.labels, total=dataset.n, desc='Statistics'):
+ x.append(np.bincount(label[:, 0].astype(int), minlength=data['nc']))
+ x = np.array(x) # shape(128x80)
+ stats[split] = {'instance_stats': {'total': int(x.sum()), 'per_class': x.sum(0).tolist()},
+ 'image_stats': {'total': dataset.n, 'unlabelled': int(np.all(x == 0, 1).sum()),
+ 'per_class': (x > 0).sum(0).tolist()},
+ 'labels': [{str(Path(k).name): round_labels(v.tolist())} for k, v in
+ zip(dataset.img_files, dataset.labels)]}
+
+ if hub:
+ im_dir = hub_dir / 'images'
+ im_dir.mkdir(parents=True, exist_ok=True)
+ for _ in tqdm(ThreadPool(NUM_THREADS).imap(hub_ops, dataset.img_files), total=dataset.n, desc='HUB Ops'):
+ pass
+
+ # Profile
+ stats_path = hub_dir / 'stats.json'
+ if profile:
+ for _ in range(1):
+ file = stats_path.with_suffix('.npy')
+ t1 = time.time()
+ np.save(file, stats)
+ t2 = time.time()
+ x = np.load(file, allow_pickle=True)
+ print(f'stats.npy times: {time.time() - t2:.3f}s read, {t2 - t1:.3f}s write')
+
+ file = stats_path.with_suffix('.json')
+ t1 = time.time()
+ with open(file, 'w') as f:
+ json.dump(stats, f) # save stats *.json
+ t2 = time.time()
+ with open(file) as f:
+ x = json.load(f) # load hyps dict
+ print(f'stats.json times: {time.time() - t2:.3f}s read, {t2 - t1:.3f}s write')
+
+ # Save, print and return
+ if hub:
+ print(f'Saving {stats_path.resolve()}...')
+ with open(stats_path, 'w') as f:
+ json.dump(stats, f) # save stats.json
+ if verbose:
+ print(json.dumps(stats, indent=2, sort_keys=False))
+ return stats
diff --git a/utils/downloads.py b/utils/downloads.py
new file mode 100644
index 0000000..a8bacae
--- /dev/null
+++ b/utils/downloads.py
@@ -0,0 +1,153 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Download utils
+"""
+
+import os
+import platform
+import subprocess
+import time
+import urllib
+from pathlib import Path
+from zipfile import ZipFile
+
+import requests
+import torch
+
+
+def gsutil_getsize(url=''):
+ # gs://bucket/file size https://cloud.google.com/storage/docs/gsutil/commands/du
+ s = subprocess.check_output(f'gsutil du {url}', shell=True).decode('utf-8')
+ return eval(s.split(' ')[0]) if len(s) else 0 # bytes
+
+
+def safe_download(file, url, url2=None, min_bytes=1E0, error_msg=''):
+ # Attempts to download file from url or url2, checks and removes incomplete downloads < min_bytes
+ file = Path(file)
+ assert_msg = f"Downloaded file '{file}' does not exist or size is < min_bytes={min_bytes}"
+ try: # url1
+ print(f'Downloading {url} to {file}...')
+ torch.hub.download_url_to_file(url, str(file))
+ assert file.exists() and file.stat().st_size > min_bytes, assert_msg # check
+ except Exception as e: # url2
+ file.unlink(missing_ok=True) # remove partial downloads
+ print(f'ERROR: {e}\nRe-attempting {url2 or url} to {file}...')
+ os.system(f"curl -L '{url2 or url}' -o '{file}' --retry 3 -C -") # curl download, retry and resume on fail
+ finally:
+ if not file.exists() or file.stat().st_size < min_bytes: # check
+ file.unlink(missing_ok=True) # remove partial downloads
+ print(f"ERROR: {assert_msg}\n{error_msg}")
+ print('')
+
+
+def attempt_download(file, repo='ultralytics/yolov5'): # from utils.downloads import *; attempt_download()
+ # Attempt file download if does not exist
+ file = Path(str(file).strip().replace("'", ''))
+
+ if not file.exists():
+ # URL specified
+ name = Path(urllib.parse.unquote(str(file))).name # decode '%2F' to '/' etc.
+ if str(file).startswith(('http:/', 'https:/')): # download
+ url = str(file).replace(':/', '://') # Pathlib turns :// -> :/
+ file = name.split('?')[0] # parse authentication https://url.com/file.txt?auth...
+ if Path(file).is_file():
+ print(f'Found {url} locally at {file}') # file already exists
+ else:
+ safe_download(file=file, url=url, min_bytes=1E5)
+ return file
+
+ # GitHub assets
+ file.parent.mkdir(parents=True, exist_ok=True) # make parent dir (if required)
+ try:
+ response = requests.get(f'https://github.com/gitapi/repos/{repo}/releases/latest').json() # github api
+ assets = [x['name'] for x in response['assets']] # release assets, i.e. ['yolov5s.pt', 'yolov5m.pt', ...]
+ tag = response['tag_name'] # i.e. 'v1.0'
+ except: # fallback plan
+ assets = ['yolov5n.pt', 'yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt',
+ 'yolov5n6.pt', 'yolov5s6.pt', 'yolov5m6.pt', 'yolov5l6.pt', 'yolov5x6.pt']
+ try:
+ tag = subprocess.check_output('git tag', shell=True, stderr=subprocess.STDOUT).decode().split()[-1]
+ except:
+ tag = 'v6.0' # current release
+
+ if name in assets:
+ safe_download(file,
+ url=f'https://github.com/{repo}/releases/download/{tag}/{name}',
+ # url2=f'https://storage.googleapis.com/{repo}/ckpt/{name}', # backup url (optional)
+ min_bytes=1E5,
+ error_msg=f'{file} missing, try downloading from https://github.com/{repo}/releases/')
+
+ return str(file)
+
+
+def gdrive_download(id='16TiPfZj7htmTyhntwcZyEEAejOUxuT6m', file='tmp.zip'):
+ # Downloads a file from Google Drive. from yolov5.utils.downloads import *; gdrive_download()
+ t = time.time()
+ file = Path(file)
+ cookie = Path('cookie') # gdrive cookie
+ print(f'Downloading https://drive.google.com/uc?export=download&id={id} as {file}... ', end='')
+ file.unlink(missing_ok=True) # remove existing file
+ cookie.unlink(missing_ok=True) # remove existing cookie
+
+ # Attempt file download
+ out = "NUL" if platform.system() == "Windows" else "/dev/null"
+ os.system(f'curl -c ./cookie -s -L "drive.google.com/uc?export=download&id={id}" > {out}')
+ if os.path.exists('cookie'): # large file
+ s = f'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm={get_token()}&id={id}" -o {file}'
+ else: # small file
+ s = f'curl -s -L -o {file} "drive.google.com/uc?export=download&id={id}"'
+ r = os.system(s) # execute, capture return
+ cookie.unlink(missing_ok=True) # remove existing cookie
+
+ # Error check
+ if r != 0:
+ file.unlink(missing_ok=True) # remove partial
+ print('Download error ') # raise Exception('Download error')
+ return r
+
+ # Unzip if archive
+ if file.suffix == '.zip':
+ print('unzipping... ', end='')
+ ZipFile(file).extractall(path=file.parent) # unzip
+ file.unlink() # remove zip
+
+ print(f'Done ({time.time() - t:.1f}s)')
+ return r
+
+
+def get_token(cookie="./cookie"):
+ with open(cookie) as f:
+ for line in f:
+ if "download" in line:
+ return line.split()[-1]
+ return ""
+
+# Google utils: https://cloud.google.com/storage/docs/reference/libraries ----------------------------------------------
+#
+#
+# def upload_blob(bucket_name, source_file_name, destination_blob_name):
+# # Uploads a file to a bucket
+# # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python
+#
+# storage_client = storage.Client()
+# bucket = storage_client.get_bucket(bucket_name)
+# blob = bucket.blob(destination_blob_name)
+#
+# blob.upload_from_filename(source_file_name)
+#
+# print('File {} uploaded to {}.'.format(
+# source_file_name,
+# destination_blob_name))
+#
+#
+# def download_blob(bucket_name, source_blob_name, destination_file_name):
+# # Uploads a blob from a bucket
+# storage_client = storage.Client()
+# bucket = storage_client.get_bucket(bucket_name)
+# blob = bucket.blob(source_blob_name)
+#
+# blob.download_to_filename(destination_file_name)
+#
+# print('Blob {} downloaded to {}.'.format(
+# source_blob_name,
+# destination_file_name))
diff --git a/utils/flask_rest_api/README.md b/utils/flask_rest_api/README.md
new file mode 100644
index 0000000..a726acb
--- /dev/null
+++ b/utils/flask_rest_api/README.md
@@ -0,0 +1,73 @@
+# Flask REST API
+
+[REST](https://en.wikipedia.org/wiki/Representational_state_transfer) [API](https://en.wikipedia.org/wiki/API)s are
+commonly used to expose Machine Learning (ML) models to other services. This folder contains an example REST API
+created using Flask to expose the YOLOv5s model from [PyTorch Hub](https://pytorch.org/hub/ultralytics_yolov5/).
+
+## Requirements
+
+[Flask](https://palletsprojects.com/p/flask/) is required. Install with:
+
+```shell
+$ pip install Flask
+```
+
+## Run
+
+After Flask installation run:
+
+```shell
+$ python3 restapi.py --port 5000
+```
+
+Then use [curl](https://curl.se/) to perform a request:
+
+```shell
+$ curl -X POST -F image=@zidane.jpg 'http://localhost:5000/v1/object-detection/yolov5s'
+```
+
+The model inference results are returned as a JSON response:
+
+```json
+[
+ {
+ "class": 0,
+ "confidence": 0.8900438547,
+ "height": 0.9318675399,
+ "name": "person",
+ "width": 0.3264600933,
+ "xcenter": 0.7438579798,
+ "ycenter": 0.5207948685
+ },
+ {
+ "class": 0,
+ "confidence": 0.8440024257,
+ "height": 0.7155083418,
+ "name": "person",
+ "width": 0.6546785235,
+ "xcenter": 0.427829951,
+ "ycenter": 0.6334488392
+ },
+ {
+ "class": 27,
+ "confidence": 0.3771208823,
+ "height": 0.3902671337,
+ "name": "tie",
+ "width": 0.0696444362,
+ "xcenter": 0.3675483763,
+ "ycenter": 0.7991207838
+ },
+ {
+ "class": 27,
+ "confidence": 0.3527112305,
+ "height": 0.1540903747,
+ "name": "tie",
+ "width": 0.0336618312,
+ "xcenter": 0.7814827561,
+ "ycenter": 0.5065554976
+ }
+]
+```
+
+An example python script to perform inference using [requests](https://docs.python-requests.org/en/master/) is given
+in `example_request.py`
diff --git a/utils/flask_rest_api/example_request.py b/utils/flask_rest_api/example_request.py
new file mode 100644
index 0000000..ff21f30
--- /dev/null
+++ b/utils/flask_rest_api/example_request.py
@@ -0,0 +1,13 @@
+"""Perform test request"""
+import pprint
+
+import requests
+
+DETECTION_URL = "http://localhost:5000/v1/object-detection/yolov5s"
+TEST_IMAGE = "zidane.jpg"
+
+image_data = open(TEST_IMAGE, "rb").read()
+
+response = requests.post(DETECTION_URL, files={"image": image_data}).json()
+
+pprint.pprint(response)
diff --git a/utils/flask_rest_api/restapi.py b/utils/flask_rest_api/restapi.py
new file mode 100644
index 0000000..b93ad16
--- /dev/null
+++ b/utils/flask_rest_api/restapi.py
@@ -0,0 +1,37 @@
+"""
+Run a rest API exposing the yolov5s object detection model
+"""
+import argparse
+import io
+
+import torch
+from flask import Flask, request
+from PIL import Image
+
+app = Flask(__name__)
+
+DETECTION_URL = "/v1/object-detection/yolov5s"
+
+
+@app.route(DETECTION_URL, methods=["POST"])
+def predict():
+ if not request.method == "POST":
+ return
+
+ if request.files.get("image"):
+ image_file = request.files["image"]
+ image_bytes = image_file.read()
+
+ img = Image.open(io.BytesIO(image_bytes))
+
+ results = model(img, size=640) # reduce size=320 for faster inference
+ return results.pandas().xyxy[0].to_json(orient="records")
+
+
+if __name__ == "__main__":
+ parser = argparse.ArgumentParser(description="Flask API exposing YOLOv5 model")
+ parser.add_argument("--port", default=5000, type=int, help="port number")
+ args = parser.parse_args()
+
+ model = torch.hub.load("ultralytics/yolov5", "yolov5s", force_reload=True) # force_reload to recache
+ app.run(host="0.0.0.0", port=args.port) # debug=True causes Restarting with stat
diff --git a/utils/general.py b/utils/general.py
new file mode 100644
index 0000000..ebc885f
--- /dev/null
+++ b/utils/general.py
@@ -0,0 +1,971 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+General utils
+"""
+
+import contextlib
+import glob
+import logging
+import math
+import os
+import platform
+import random
+import re
+import shutil
+import signal
+import time
+import urllib
+from itertools import repeat
+from multiprocessing.pool import ThreadPool
+from pathlib import Path
+from subprocess import check_output
+from zipfile import ZipFile
+
+import cv2
+import numpy as np
+import pandas as pd
+import pkg_resources as pkg
+import torch
+import torchvision
+import yaml
+
+from utils.downloads import gsutil_getsize
+from utils.metrics import box_iou, fitness
+pi = 3.141592
+from utils.nms_rotated import obb_nms
+
+# Settings
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[1] # YOLOv5 root directory
+NUM_THREADS = min(8, max(1, os.cpu_count() - 1)) # number of YOLOv5 multiprocessing threads
+
+torch.set_printoptions(linewidth=320, precision=5, profile='long')
+np.set_printoptions(linewidth=320, formatter={'float_kind': '{:11.5g}'.format}) # format short g, %precision=5
+pd.options.display.max_columns = 10
+cv2.setNumThreads(0) # prevent OpenCV from multithreading (incompatible with PyTorch DataLoader)
+os.environ['NUMEXPR_MAX_THREADS'] = str(NUM_THREADS) # NumExpr max threads
+
+
+def set_logging(name=None, verbose=True):
+ # Sets level and returns logger
+ for h in logging.root.handlers:
+ logging.root.removeHandler(h) # remove all handlers associated with the root logger object
+ rank = int(os.getenv('RANK', -1)) # rank in world for Multi-GPU trainings
+ logging.basicConfig(format="%(message)s", level=logging.INFO if (verbose and rank in (-1, 0)) else logging.WARNING)
+ return logging.getLogger(name)
+
+
+LOGGER = set_logging(__name__) # define globally (used in train.py, val.py, detect.py, etc.)
+
+
+class Profile(contextlib.ContextDecorator):
+ # Usage: @Profile() decorator or 'with Profile():' context manager
+ def __enter__(self):
+ self.start = time.time()
+
+ def __exit__(self, type, value, traceback):
+ print(f'Profile results: {time.time() - self.start:.5f}s')
+
+
+class Timeout(contextlib.ContextDecorator):
+ # Usage: @Timeout(seconds) decorator or 'with Timeout(seconds):' context manager
+ def __init__(self, seconds, *, timeout_msg='', suppress_timeout_errors=True):
+ self.seconds = int(seconds)
+ self.timeout_message = timeout_msg
+ self.suppress = bool(suppress_timeout_errors)
+
+ def _timeout_handler(self, signum, frame):
+ raise TimeoutError(self.timeout_message)
+
+ def __enter__(self):
+ signal.signal(signal.SIGALRM, self._timeout_handler) # Set handler for SIGALRM
+ signal.alarm(self.seconds) # start countdown for SIGALRM to be raised
+
+ def __exit__(self, exc_type, exc_val, exc_tb):
+ signal.alarm(0) # Cancel SIGALRM if it's scheduled
+ if self.suppress and exc_type is TimeoutError: # Suppress TimeoutError
+ return True
+
+
+class WorkingDirectory(contextlib.ContextDecorator):
+ # Usage: @WorkingDirectory(dir) decorator or 'with WorkingDirectory(dir):' context manager
+ def __init__(self, new_dir):
+ self.dir = new_dir # new dir
+ self.cwd = Path.cwd().resolve() # current dir
+
+ def __enter__(self):
+ os.chdir(self.dir)
+
+ def __exit__(self, exc_type, exc_val, exc_tb):
+ os.chdir(self.cwd)
+
+
+def try_except(func):
+ # try-except function. Usage: @try_except decorator
+ def handler(*args, **kwargs):
+ try:
+ func(*args, **kwargs)
+ except Exception as e:
+ print(e)
+
+ return handler
+
+
+def methods(instance):
+ # Get class/instance methods
+ return [f for f in dir(instance) if callable(getattr(instance, f)) and not f.startswith("__")]
+
+
+def print_args(name, opt):
+ # Print argparser arguments
+ LOGGER.info(colorstr(f'{name}: ') + ', '.join(f'{k}={v}' for k, v in vars(opt).items()))
+
+
+def init_seeds(seed=0):
+ # Initialize random number generator (RNG) seeds https://pytorch.org/docs/stable/notes/randomness.html
+ # cudnn seed 0 settings are slower and more reproducible, else faster and less reproducible
+ import torch.backends.cudnn as cudnn
+ random.seed(seed)
+ np.random.seed(seed)
+ torch.manual_seed(seed)
+ cudnn.benchmark, cudnn.deterministic = (False, True) if seed == 0 else (True, False)
+
+
+def intersect_dicts(da, db, exclude=()):
+ # Dictionary intersection of matching keys and shapes, omitting 'exclude' keys, using da values
+ return {k: v for k, v in da.items() if k in db and not any(x in k for x in exclude) and v.shape == db[k].shape}
+
+
+def get_latest_run(search_dir='.'):
+ # Return path to most recent 'last.pt' in /runs (i.e. to --resume from)
+ last_list = glob.glob(f'{search_dir}/**/last*.pt', recursive=True)
+ return max(last_list, key=os.path.getctime) if last_list else ''
+
+
+def user_config_dir(dir='Ultralytics', env_var='YOLOV5_CONFIG_DIR'):
+ # Return path of user configuration directory. Prefer environment variable if exists. Make dir if required.
+ env = os.getenv(env_var)
+ if env:
+ path = Path(env) # use environment variable
+ else:
+ cfg = {'Windows': 'AppData/Roaming', 'Linux': '.config', 'Darwin': 'Library/Application Support'} # 3 OS dirs
+ path = Path.home() / cfg.get(platform.system(), '') # OS-specific config dir
+ path = (path if is_writeable(path) else Path('/tmp')) / dir # GCP and AWS lambda fix, only /tmp is writeable
+ path.mkdir(exist_ok=True) # make if required
+ return path
+
+
+def is_writeable(dir, test=False):
+ # Return True if directory has write permissions, test opening a file with write permissions if test=True
+ if test: # method 1
+ file = Path(dir) / 'tmp.txt'
+ try:
+ with open(file, 'w'): # open file with write permissions
+ pass
+ file.unlink() # remove file
+ return True
+ except OSError:
+ return False
+ else: # method 2
+ return os.access(dir, os.R_OK) # possible issues on Windows
+
+
+def is_docker():
+ # Is environment a Docker container?
+ return Path('/workspace').exists() # or Path('/.dockerenv').exists()
+
+
+def is_colab():
+ # Is environment a Google Colab instance?
+ try:
+ import google.colab
+ return True
+ except ImportError:
+ return False
+
+
+def is_pip():
+ # Is file in a pip package?
+ return 'site-packages' in Path(__file__).resolve().parts
+
+
+def is_ascii(s=''):
+ # Is string composed of all ASCII (no UTF) characters? (note str().isascii() introduced in python 3.7)
+ s = str(s) # convert list, tuple, None, etc. to str
+ return len(s.encode().decode('ascii', 'ignore')) == len(s)
+
+
+def is_chinese(s='人工智能'):
+ # Is string composed of any Chinese characters?
+ return re.search('[\u4e00-\u9fff]', s)
+
+
+def emojis(str=''):
+ # Return platform-dependent emoji-safe version of string
+ return str.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else str
+
+
+def file_size(path):
+ # Return file/dir size (MB)
+ path = Path(path)
+ if path.is_file():
+ return path.stat().st_size / 1E6
+ elif path.is_dir():
+ return sum(f.stat().st_size for f in path.glob('**/*') if f.is_file()) / 1E6
+ else:
+ return 0.0
+
+
+def check_online():
+ # Check internet connectivity
+ import socket
+ try:
+ socket.create_connection(("1.1.1.1", 443), 5) # check host accessibility
+ return True
+ except OSError:
+ return False
+
+
+@try_except
+@WorkingDirectory(ROOT)
+def check_git_status():
+ # Recommend 'git pull' if code is out of date
+ msg = ', for updates see https://github.com/ultralytics/yolov5'
+ print(colorstr('github: '), end='')
+ assert Path('.git').exists(), 'skipping check (not a git repository)' + msg
+ assert not is_docker(), 'skipping check (Docker image)' + msg
+ assert check_online(), 'skipping check (offline)' + msg
+
+ cmd = 'git fetch && git config --get remote.origin.url'
+ url = check_output(cmd, shell=True, timeout=5).decode().strip().rstrip('.git') # git fetch
+ branch = check_output('git rev-parse --abbrev-ref HEAD', shell=True).decode().strip() # checked out
+ n = int(check_output(f'git rev-list {branch}..origin/master --count', shell=True)) # commits behind
+ if n > 0:
+ s = f"⚠️ YOLOv5 is out of date by {n} commit{'s' * (n > 1)}. Use `git pull` or `git clone {url}` to update."
+ else:
+ s = f'up to date with {url} ✅'
+ print(emojis(s)) # emoji-safe
+
+
+def check_python(minimum='3.6.2'):
+ # Check current python version vs. required python version
+ check_version(platform.python_version(), minimum, name='Python ', hard=True)
+
+
+def check_version(current='0.0.0', minimum='0.0.0', name='version ', pinned=False, hard=False, verbose=False):
+ # Check version vs. required version
+ current, minimum = (pkg.parse_version(x) for x in (current, minimum))
+ result = (current == minimum) if pinned else (current >= minimum) # bool
+ s = f'{name}{minimum} required by YOLOv5, but {name}{current} is currently installed' # string
+ if hard:
+ assert result, s # assert min requirements met
+ if verbose and not result:
+ LOGGER.warning(s)
+ return result
+
+
+@try_except
+def check_requirements(requirements=ROOT / 'requirements.txt', exclude=(), install=True):
+ # Check installed dependencies meet requirements (pass *.txt file or list of packages)
+ prefix = colorstr('red', 'bold', 'requirements:')
+ check_python() # check python version
+ if isinstance(requirements, (str, Path)): # requirements.txt file
+ file = Path(requirements)
+ assert file.exists(), f"{prefix} {file.resolve()} not found, check failed."
+ with file.open() as f:
+ requirements = [f'{x.name}{x.specifier}' for x in pkg.parse_requirements(f) if x.name not in exclude]
+ else: # list or tuple of packages
+ requirements = [x for x in requirements if x not in exclude]
+
+ n = 0 # number of packages updates
+ for r in requirements:
+ try:
+ pkg.require(r)
+ except Exception as e: # DistributionNotFound or VersionConflict if requirements not met
+ s = f"{prefix} {r} not found and is required by YOLOv5"
+ if install:
+ print(f"{s}, attempting auto-update...")
+ try:
+ assert check_online(), f"'pip install {r}' skipped (offline)"
+ print(check_output(f"pip install '{r}'", shell=True).decode())
+ n += 1
+ except Exception as e:
+ print(f'{prefix} {e}')
+ else:
+ print(f'{s}. Please install and rerun your command.')
+
+ if n: # if packages updated
+ source = file.resolve() if 'file' in locals() else requirements
+ s = f"{prefix} {n} package{'s' * (n > 1)} updated per {source}\n" \
+ f"{prefix} ⚠️ {colorstr('bold', 'Restart runtime or rerun command for updates to take effect')}\n"
+ print(emojis(s))
+
+
+def check_img_size(imgsz, s=32, floor=0):
+ # Verify image size is a multiple of stride s in each dimension
+ if isinstance(imgsz, int): # integer i.e. img_size=640
+ new_size = max(make_divisible(imgsz, int(s)), floor)
+ else: # list i.e. img_size=[640, 480]
+ new_size = [max(make_divisible(x, int(s)), floor) for x in imgsz]
+ if new_size != imgsz:
+ print(f'WARNING: --img-size {imgsz} must be multiple of max stride {s}, updating to {new_size}')
+ return new_size
+
+
+def check_imshow():
+ # Check if environment supports image displays
+ try:
+ assert not is_docker(), 'cv2.imshow() is disabled in Docker environments'
+ assert not is_colab(), 'cv2.imshow() is disabled in Google Colab environments'
+ cv2.imshow('test', np.zeros((1, 1, 3)))
+ cv2.waitKey(1)
+ cv2.destroyAllWindows()
+ cv2.waitKey(1)
+ return True
+ except Exception as e:
+ print(f'WARNING: Environment does not support cv2.imshow() or PIL Image.show() image displays\n{e}')
+ return False
+
+
+def check_suffix(file='yolov5s.pt', suffix=('.pt',), msg=''):
+ # Check file(s) for acceptable suffix
+ if file and suffix:
+ if isinstance(suffix, str):
+ suffix = [suffix]
+ for f in file if isinstance(file, (list, tuple)) else [file]:
+ s = Path(f).suffix.lower() # file suffix
+ if len(s):
+ assert s in suffix, f"{msg}{f} acceptable suffix is {suffix}"
+
+
+def check_yaml(file, suffix=('.yaml', '.yml')):
+ # Search/download YAML file (if necessary) and return path, checking suffix
+ return check_file(file, suffix)
+
+
+def check_file(file, suffix=''):
+ # Search/download file (if necessary) and return path
+ check_suffix(file, suffix) # optional
+ file = str(file) # convert to str()
+ if Path(file).is_file() or file == '': # exists
+ return file
+ elif file.startswith(('http:/', 'https:/')): # download
+ url = str(Path(file)).replace(':/', '://') # Pathlib turns :// -> :/
+ file = Path(urllib.parse.unquote(file).split('?')[0]).name # '%2F' to '/', split https://url.com/file.txt?auth
+ if Path(file).is_file():
+ print(f'Found {url} locally at {file}') # file already exists
+ else:
+ print(f'Downloading {url} to {file}...')
+ torch.hub.download_url_to_file(url, file)
+ assert Path(file).exists() and Path(file).stat().st_size > 0, f'File download failed: {url}' # check
+ return file
+ else: # search
+ files = []
+ for d in 'data', 'models', 'utils': # search directories
+ files.extend(glob.glob(str(ROOT / d / '**' / file), recursive=True)) # find file
+ assert len(files), f'File not found: {file}' # assert file was found
+ assert len(files) == 1, f"Multiple files match '{file}', specify exact path: {files}" # assert unique
+ return files[0] # return file
+
+
+def check_dataset(data, autodownload=True):
+ # Download and/or unzip dataset if not found locally
+ # Usage: https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128_with_yaml.zip
+
+ # Download (optional)
+ extract_dir = ''
+ if isinstance(data, (str, Path)) and str(data).endswith('.zip'): # i.e. gs://bucket/dir/coco128.zip
+ download(data, dir='../datasets', unzip=True, delete=False, curl=False, threads=1)
+ data = next((Path('../datasets') / Path(data).stem).rglob('*.yaml'))
+ extract_dir, autodownload = data.parent, False
+
+ # Read yaml (optional)
+ if isinstance(data, (str, Path)):
+ with open(data, errors='ignore') as f:
+ data = yaml.safe_load(f) # dictionary
+
+ # Parse yaml
+ path = extract_dir or Path(data.get('path') or '') # optional 'path' default to '.'
+ for k in 'train', 'val', 'test':
+ if data.get(k): # prepend path
+ data[k] = str(path / data[k]) if isinstance(data[k], str) else [str(path / x) for x in data[k]]
+
+ assert 'nc' in data, "Dataset 'nc' key missing."
+ if 'names' not in data:
+ data['names'] = [f'class{i}' for i in range(data['nc'])] # assign class names if missing
+ train, val, test, s = (data.get(x) for x in ('train', 'val', 'test', 'download'))
+ if val:
+ val = [Path(x).resolve() for x in (val if isinstance(val, list) else [val])] # val path
+ if not all(x.exists() for x in val):
+ print('\nWARNING: Dataset not found, nonexistent paths: %s' % [str(x) for x in val if not x.exists()])
+ if s and autodownload: # download script
+ root = path.parent if 'path' in data else '..' # unzip directory i.e. '../'
+ if s.startswith('http') and s.endswith('.zip'): # URL
+ f = Path(s).name # filename
+ print(f'Downloading {s} to {f}...')
+ torch.hub.download_url_to_file(s, f)
+ Path(root).mkdir(parents=True, exist_ok=True) # create root
+ ZipFile(f).extractall(path=root) # unzip
+ Path(f).unlink() # remove zip
+ r = None # success
+ elif s.startswith('bash '): # bash script
+ print(f'Running {s} ...')
+ r = os.system(s)
+ else: # python script
+ r = exec(s, {'yaml': data}) # return None
+ print(f"Dataset autodownload {f'success, saved to {root}' if r in (0, None) else 'failure'}\n")
+ else:
+ raise Exception('Dataset not found.')
+
+ return data # dictionary
+
+
+def url2file(url):
+ # Convert URL to filename, i.e. https://url.com/file.txt?auth -> file.txt
+ url = str(Path(url)).replace(':/', '://') # Pathlib turns :// -> :/
+ file = Path(urllib.parse.unquote(url)).name.split('?')[0] # '%2F' to '/', split https://url.com/file.txt?auth
+ return file
+
+
+def download(url, dir='.', unzip=True, delete=True, curl=False, threads=1):
+ # Multi-threaded file download and unzip function, used in data.yaml for autodownload
+ def download_one(url, dir):
+ # Download 1 file
+ f = dir / Path(url).name # filename
+ if Path(url).is_file(): # exists in current path
+ Path(url).rename(f) # move to dir
+ elif not f.exists():
+ print(f'Downloading {url} to {f}...')
+ if curl:
+ os.system(f"curl -L '{url}' -o '{f}' --retry 9 -C -") # curl download, retry and resume on fail
+ else:
+ torch.hub.download_url_to_file(url, f, progress=True) # torch download
+ if unzip and f.suffix in ('.zip', '.gz'):
+ print(f'Unzipping {f}...')
+ if f.suffix == '.zip':
+ ZipFile(f).extractall(path=dir) # unzip
+ elif f.suffix == '.gz':
+ os.system(f'tar xfz {f} --directory {f.parent}') # unzip
+ if delete:
+ f.unlink() # remove zip
+
+ dir = Path(dir)
+ dir.mkdir(parents=True, exist_ok=True) # make directory
+ if threads > 1:
+ pool = ThreadPool(threads)
+ pool.imap(lambda x: download_one(*x), zip(url, repeat(dir))) # multi-threaded
+ pool.close()
+ pool.join()
+ else:
+ for u in [url] if isinstance(url, (str, Path)) else url:
+ download_one(u, dir)
+
+
+def make_divisible(x, divisor):
+ # Returns nearest x divisible by divisor
+ if isinstance(divisor, torch.Tensor):
+ divisor = int(divisor.max()) # to int
+ return math.ceil(x / divisor) * divisor
+
+
+def clean_str(s):
+ # Cleans a string by replacing special characters with underscore _
+ return re.sub(pattern="[|@#!¡·$€%&()=?¿^*;:,¨´><+]", repl="_", string=s)
+
+
+def one_cycle(y1=0.0, y2=1.0, steps=100):
+ # lambda function for sinusoidal ramp from y1 to y2 https://arxiv.org/pdf/1812.01187.pdf
+ return lambda x: ((1 - math.cos(x * math.pi / steps)) / 2) * (y2 - y1) + y1
+
+
+def colorstr(*input):
+ # Colors a string https://en.wikipedia.org/wiki/ANSI_escape_code, i.e. colorstr('blue', 'hello world')
+ *args, string = input if len(input) > 1 else ('blue', 'bold', input[0]) # color arguments, string
+ colors = {'black': '\033[30m', # basic colors
+ 'red': '\033[31m',
+ 'green': '\033[32m',
+ 'yellow': '\033[33m',
+ 'blue': '\033[34m',
+ 'magenta': '\033[35m',
+ 'cyan': '\033[36m',
+ 'white': '\033[37m',
+ 'bright_black': '\033[90m', # bright colors
+ 'bright_red': '\033[91m',
+ 'bright_green': '\033[92m',
+ 'bright_yellow': '\033[93m',
+ 'bright_blue': '\033[94m',
+ 'bright_magenta': '\033[95m',
+ 'bright_cyan': '\033[96m',
+ 'bright_white': '\033[97m',
+ 'end': '\033[0m', # misc
+ 'bold': '\033[1m',
+ 'underline': '\033[4m'}
+ return ''.join(colors[x] for x in args) + f'{string}' + colors['end']
+
+
+def labels_to_class_weights(labels, nc=80):
+ # Get class weights (inverse frequency) from training labels
+ if labels[0] is None: # no labels loaded
+ return torch.Tensor()
+
+ labels = np.concatenate(labels, 0) # labels.shape = (866643, 5) for COCO
+ classes = labels[:, 0].astype(np.int) # labels = [class xywh]
+ weights = np.bincount(classes, minlength=nc) # occurrences per class
+
+ # Prepend gridpoint count (for uCE training)
+ # gpi = ((320 / 32 * np.array([1, 2, 4])) ** 2 * 3).sum() # gridpoints per image
+ # weights = np.hstack([gpi * len(labels) - weights.sum() * 9, weights * 9]) ** 0.5 # prepend gridpoints to start
+
+ weights[weights == 0] = 1 # replace empty bins with 1
+ weights = 1 / weights # number of targets per class
+ weights /= weights.sum() # normalize
+ return torch.from_numpy(weights)
+
+
+def labels_to_image_weights(labels, nc=80, class_weights=np.ones(80)):
+ # Produces image weights based on class_weights and image contents
+ class_counts = np.array([np.bincount(x[:, 0].astype(np.int), minlength=nc) for x in labels])
+ image_weights = (class_weights.reshape(1, nc) * class_counts).sum(1)
+ # index = random.choices(range(n), weights=image_weights, k=1) # weight image sample
+ return image_weights
+
+
+def coco80_to_coco91_class(): # converts 80-index (val2014) to 91-index (paper)
+ # https://tech.amikelive.com/node-718/what-object-categories-labels-are-in-coco-dataset/
+ # a = np.loadtxt('data/coco.names', dtype='str', delimiter='\n')
+ # b = np.loadtxt('data/coco_paper.names', dtype='str', delimiter='\n')
+ # x1 = [list(a[i] == b).index(True) + 1 for i in range(80)] # darknet to coco
+ # x2 = [list(b[i] == a).index(True) if any(b[i] == a) else None for i in range(91)] # coco to darknet
+ x = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27, 28, 31, 32, 33, 34,
+ 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63,
+ 64, 65, 67, 70, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 84, 85, 86, 87, 88, 89, 90]
+ return x
+
+
+def xyxy2xywh(x):
+ # Convert nx4 boxes from [x1, y1, x2, y2] to [x, y, w, h] where xy1=top-left, xy2=bottom-right
+ y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+ y[:, 0] = (x[:, 0] + x[:, 2]) / 2 # x center
+ y[:, 1] = (x[:, 1] + x[:, 3]) / 2 # y center
+ y[:, 2] = x[:, 2] - x[:, 0] # width
+ y[:, 3] = x[:, 3] - x[:, 1] # height
+ return y
+
+
+def xywh2xyxy(x):
+ # Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
+ y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+ y[:, 0] = x[:, 0] - x[:, 2] / 2 # top left x
+ y[:, 1] = x[:, 1] - x[:, 3] / 2 # top left y
+ y[:, 2] = x[:, 0] + x[:, 2] / 2 # bottom right x
+ y[:, 3] = x[:, 1] + x[:, 3] / 2 # bottom right y
+ return y
+
+
+def xywhn2xyxy(x, w=640, h=640, padw=0, padh=0):
+ # Convert nx4 boxes from [x, y, w, h] normalized to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
+ y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+ y[:, 0] = w * (x[:, 0] - x[:, 2] / 2) + padw # top left x
+ y[:, 1] = h * (x[:, 1] - x[:, 3] / 2) + padh # top left y
+ y[:, 2] = w * (x[:, 0] + x[:, 2] / 2) + padw # bottom right x
+ y[:, 3] = h * (x[:, 1] + x[:, 3] / 2) + padh # bottom right y
+ return y
+
+
+def xyxy2xywhn(x, w=640, h=640, clip=False, eps=0.0):
+ # Convert nx4 boxes from [x1, y1, x2, y2] to [x, y, w, h] normalized where xy1=top-left, xy2=bottom-right
+ if clip:
+ clip_coords(x, (h - eps, w - eps)) # warning: inplace clip
+ y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+ y[:, 0] = ((x[:, 0] + x[:, 2]) / 2) / w # x center
+ y[:, 1] = ((x[:, 1] + x[:, 3]) / 2) / h # y center
+ y[:, 2] = (x[:, 2] - x[:, 0]) / w # width
+ y[:, 3] = (x[:, 3] - x[:, 1]) / h # height
+ return y
+
+
+def xyn2xy(x, w=640, h=640, padw=0, padh=0):
+ # Convert normalized segments into pixel segments, shape (n,2)
+ y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
+ y[:, 0] = w * x[:, 0] + padw # top left x
+ y[:, 1] = h * x[:, 1] + padh # top left y
+ return y
+
+
+def segment2box(segment, width=640, height=640):
+ # Convert 1 segment label to 1 box label, applying inside-image constraint, i.e. (xy1, xy2, ...) to (xyxy)
+ x, y = segment.T # segment xy
+ inside = (x >= 0) & (y >= 0) & (x <= width) & (y <= height)
+ x, y, = x[inside], y[inside]
+ return np.array([x.min(), y.min(), x.max(), y.max()]) if any(x) else np.zeros((1, 4)) # xyxy
+
+
+def segments2boxes(segments):
+ # Convert segment labels to box labels, i.e. (cls, xy1, xy2, ...) to (cls, xywh)
+ boxes = []
+ for s in segments:
+ x, y = s.T # segment xy
+ boxes.append([x.min(), y.min(), x.max(), y.max()]) # cls, xyxy
+ return xyxy2xywh(np.array(boxes)) # cls, xywh
+
+
+def resample_segments(segments, n=1000):
+ # Up-sample an (n,2) segment
+ for i, s in enumerate(segments):
+ x = np.linspace(0, len(s) - 1, n)
+ xp = np.arange(len(s))
+ segments[i] = np.concatenate([np.interp(x, xp, s[:, i]) for i in range(2)]).reshape(2, -1).T # segment xy
+ return segments
+
+
+def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
+ # Rescale coords (xyxy) from img1_shape to img0_shape
+ if ratio_pad is None: # calculate from img0_shape
+ gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1]) # gain = old / new
+ pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2 # wh padding
+ else:
+ gain = ratio_pad[0][0]
+ pad = ratio_pad[1]
+
+ coords[:, [0, 2]] -= pad[0] # x padding
+ coords[:, [1, 3]] -= pad[1] # y padding
+ coords[:, :4] /= gain
+ clip_coords(coords, img0_shape)
+ return coords
+
+def scale_polys(img1_shape, polys, img0_shape, ratio_pad=None):
+ # ratio_pad: [(h_raw, w_raw), (hw_ratios, wh_paddings)]
+ # Rescale coords (xyxyxyxy) from img1_shape to img0_shape
+ if ratio_pad is None: # calculate from img0_shape
+ gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1]) # gain = resized / raw
+ pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2 # wh padding
+ else:
+ gain = ratio_pad[0][0] # h_ratios
+ pad = ratio_pad[1] # wh_paddings
+
+ polys[:, [0, 2, 4, 6]] -= pad[0] # x padding
+ polys[:, [1, 3, 5, 7]] -= pad[1] # y padding
+ polys[:, :8] /= gain # Rescale poly shape to img0_shape
+ #clip_polys(polys, img0_shape)
+ return polys
+
+def clip_polys(polys, shape):
+ # Clip bounding xyxyxyxy bounding boxes to image shape (height, width)
+ if isinstance(polys, torch.Tensor): # faster individually
+ polys[:, 0].clamp_(0, shape[1]) # x1
+ polys[:, 1].clamp_(0, shape[0]) # y1
+ polys[:, 2].clamp_(0, shape[1]) # x2
+ polys[:, 3].clamp_(0, shape[0]) # y2
+ polys[:, 4].clamp_(0, shape[1]) # x3
+ polys[:, 5].clamp_(0, shape[0]) # y3
+ polys[:, 6].clamp_(0, shape[1]) # x4
+ polys[:, 7].clamp_(0, shape[0]) # y4
+ else: # np.array (faster grouped)
+ polys[:, [0, 2, 4, 6]] = polys[:, [0, 2, 4, 6]].clip(0, shape[1]) # x1, x2, x3, x4
+ polys[:, [1, 3, 5, 7]] = polys[:, [1, 3, 5, 7]].clip(0, shape[0]) # y1, y2, y3, y4
+
+def clip_coords(boxes, shape):
+ # Clip bounding xyxy bounding boxes to image shape (height, width)
+ if isinstance(boxes, torch.Tensor): # faster individually
+ boxes[:, 0].clamp_(0, shape[1]) # x1
+ boxes[:, 1].clamp_(0, shape[0]) # y1
+ boxes[:, 2].clamp_(0, shape[1]) # x2
+ boxes[:, 3].clamp_(0, shape[0]) # y2
+ else: # np.array (faster grouped)
+ boxes[:, [0, 2]] = boxes[:, [0, 2]].clip(0, shape[1]) # x1, x2
+ boxes[:, [1, 3]] = boxes[:, [1, 3]].clip(0, shape[0]) # y1, y2
+
+
+def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, multi_label=False,
+ labels=(), max_det=300):
+ """Runs Non-Maximum Suppression (NMS) on inference results
+
+ Returns:
+ list of detections, on (n,6) tensor per image [xyxy, conf, cls]
+ """
+
+ nc = prediction.shape[2] - 5 # number of classes
+ xc = prediction[..., 4] > conf_thres # candidates
+
+ # Checks
+ assert 0 <= conf_thres <= 1, f'Invalid Confidence threshold {conf_thres}, valid values are between 0.0 and 1.0'
+ assert 0 <= iou_thres <= 1, f'Invalid IoU {iou_thres}, valid values are between 0.0 and 1.0'
+
+ # Settings
+ min_wh, max_wh = 2, 4096 # (pixels) minimum and maximum box width and height
+ max_nms = 30000 # maximum number of boxes into torchvision.ops.nms()
+ time_limit = 10.0 # seconds to quit after
+ redundant = True # require redundant detections
+ multi_label &= nc > 1 # multiple labels per box (adds 0.5ms/img)
+ merge = False # use merge-NMS
+
+ t = time.time()
+ output = [torch.zeros((0, 6), device=prediction.device)] * prediction.shape[0]
+ for xi, x in enumerate(prediction): # image index, image inference
+ # Apply constraints
+ # x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0 # width-height
+ x = x[xc[xi]] # confidence
+
+ # Cat apriori labels if autolabelling
+ if labels and len(labels[xi]):
+ l = labels[xi]
+ v = torch.zeros((len(l), nc + 5), device=x.device)
+ v[:, :4] = l[:, 1:5] # box
+ v[:, 4] = 1.0 # conf
+ v[range(len(l)), l[:, 0].long() + 5] = 1.0 # cls
+ x = torch.cat((x, v), 0)
+
+ # If none remain process next image
+ if not x.shape[0]:
+ continue
+
+ # Compute conf
+ x[:, 5:] *= x[:, 4:5] # conf = obj_conf * cls_conf
+
+ # Box (center x, center y, width, height) to (x1, y1, x2, y2)
+ box = xywh2xyxy(x[:, :4])
+
+ # Detections matrix nx6 (xyxy, conf, cls)
+ if multi_label:
+ i, j = (x[:, 5:] > conf_thres).nonzero(as_tuple=False).T
+ x = torch.cat((box[i], x[i, j + 5, None], j[:, None].float()), 1)
+ else: # best class only
+ conf, j = x[:, 5:].max(1, keepdim=True)
+ x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]
+
+ # Filter by class
+ if classes is not None:
+ x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]
+
+ # Apply finite constraint
+ # if not torch.isfinite(x).all():
+ # x = x[torch.isfinite(x).all(1)]
+
+ # Check shape
+ n = x.shape[0] # number of boxes
+ if not n: # no boxes
+ continue
+ elif n > max_nms: # excess boxes
+ x = x[x[:, 4].argsort(descending=True)[:max_nms]] # sort by confidence
+
+ # Batched NMS
+ c = x[:, 5:6] * (0 if agnostic else max_wh) # classes
+ boxes, scores = x[:, :4] + c, x[:, 4] # boxes (offset by class), scores
+ i = torchvision.ops.nms(boxes, scores, iou_thres) # NMS
+ if i.shape[0] > max_det: # limit detections
+ i = i[:max_det]
+ if merge and (1 < n < 3E3): # Merge NMS (boxes merged using weighted mean)
+ # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
+ iou = box_iou(boxes[i], boxes) > iou_thres # iou matrix
+ weights = iou * scores[None] # box weights
+ x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True) # merged boxes
+ if redundant:
+ i = i[iou.sum(1) > 1] # require redundancy
+
+ output[xi] = x[i]
+ if (time.time() - t) > time_limit:
+ print(f'WARNING: NMS time limit {time_limit}s exceeded')
+ break # time limit exceeded
+
+ return output
+
+def non_max_suppression_obb(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, multi_label=False,
+ labels=(), max_det=1500):
+ """Runs Non-Maximum Suppression (NMS) on inference results_obb
+ Args:
+ prediction (tensor): (b, n_all_anchors, [cx cy l s obj num_cls theta_cls])
+ agnostic (bool): True = NMS will be applied between elements of different categories
+ labels : () or
+
+ Returns:
+ list of detections, len=batch_size, on (n,7) tensor per image [xylsθ, conf, cls] θ ∈ [-pi/2, pi/2)
+ """
+
+ nc = prediction.shape[2] - 5 - 180 # number of classes
+ xc = prediction[..., 4] > conf_thres # candidates
+ class_index = nc + 5
+
+ # Checks
+ assert 0 <= conf_thres <= 1, f'Invalid Confidence threshold {conf_thres}, valid values are between 0.0 and 1.0'
+ assert 0 <= iou_thres <= 1, f'Invalid IoU {iou_thres}, valid values are between 0.0 and 1.0'
+
+ # Settings
+ max_wh = 4096 # min_wh, max_wh = 2, 4096 # (pixels) minimum and maximum box width and height
+ max_nms = 30000 # maximum number of boxes into torchvision.ops.nms()
+ time_limit = 30.0 # seconds to quit after
+ # redundant = True # require redundant detections
+ multi_label &= nc > 1 # multiple labels per box (adds 0.5ms/img)
+
+ t = time.time()
+ output = [torch.zeros((0, 7), device=prediction.device)] * prediction.shape[0]
+ for xi, x in enumerate(prediction): # image index, image inference
+ # Apply constraints
+ # x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0 # width-height
+ x = x[xc[xi]] # confidence, (tensor): (n_conf_thres, [cx cy l s obj num_cls theta_cls])
+
+ # Cat apriori labels if autolabelling
+ if labels and len(labels[xi]):
+ l = labels[xi]
+ v = torch.zeros((len(l), nc + 5), device=x.device)
+ v[:, :4] = l[:, 1:5] # box
+ v[:, 4] = 1.0 # conf
+ v[range(len(l)), l[:, 0].long() + 5] = 1.0 # cls
+ x = torch.cat((x, v), 0)
+
+ # If none remain process next image
+ if not x.shape[0]:
+ continue
+
+ # Compute conf
+ x[:, 5:class_index] *= x[:, 4:5] # conf = obj_conf * cls_conf
+
+ _, theta_pred = torch.max(x[:, class_index:], 1, keepdim=True) # [n_conf_thres, 1] θ ∈ int[0, 179]
+ theta_pred = (theta_pred - 90) / 180 * pi # [n_conf_thres, 1] θ ∈ [-pi/2, pi/2)
+
+ # Detections matrix nx7 (xyls, θ, conf, cls) θ ∈ [-pi/2, pi/2)
+ if multi_label:
+ i, j = (x[:, 5:class_index] > conf_thres).nonzero(as_tuple=False).T # ()
+ x = torch.cat((x[i, :4], theta_pred[i], x[i, j + 5, None], j[:, None].float()), 1)
+ else: # best class only
+ conf, j = x[:, 5:class_index].max(1, keepdim=True)
+ x = torch.cat((x[:, :4], theta_pred, conf, j.float()), 1)[conf.view(-1) > conf_thres]
+
+ # Filter by class
+ if classes is not None:
+ x = x[(x[:, 6:7] == torch.tensor(classes, device=x.device)).any(1)]
+
+ # Apply finite constraint
+ # if not torch.isfinite(x).all():
+ # x = x[torch.isfinite(x).all(1)]
+
+ # Check shape
+ n = x.shape[0] # number of boxes
+ if not n: # no boxes
+ continue
+ elif n > max_nms: # excess boxes
+ x = x[x[:, 5].argsort(descending=True)[:max_nms]] # sort by confidence
+
+ # Batched NMS
+ c = x[:, 6:7] * (0 if agnostic else max_wh) # classes
+ rboxes = x[:, :5].clone()
+ rboxes[:, :2] = rboxes[:, :2] + c # rboxes (offset by class)
+ scores = x[:, 5] # scores
+ _, i = obb_nms(rboxes, scores, iou_thres)
+ if i.shape[0] > max_det: # limit detections
+ i = i[:max_det]
+
+ output[xi] = x[i]
+ if (time.time() - t) > time_limit:
+ print(f'WARNING: NMS time limit {time_limit}s exceeded')
+ break # time limit exceeded
+
+ return output
+
+def strip_optimizer(f='best.pt', s=''): # from utils.general import *; strip_optimizer()
+ # Strip optimizer from 'f' to finalize training, optionally save as 's'
+ x = torch.load(f, map_location=torch.device('cpu'))
+ if x.get('ema'):
+ x['model'] = x['ema'] # replace model with ema
+ for k in 'optimizer', 'best_fitness', 'wandb_id', 'ema', 'updates': # keys
+ x[k] = None
+ x['epoch'] = -1
+ x['model'].half() # to FP16
+ for p in x['model'].parameters():
+ p.requires_grad = False
+ torch.save(x, s or f)
+ mb = os.path.getsize(s or f) / 1E6 # filesize
+ print(f"Optimizer stripped from {f},{(' saved as %s,' % s) if s else ''} {mb:.1f}MB")
+
+
+def print_mutation(results, hyp, save_dir, bucket):
+ evolve_csv, results_csv, evolve_yaml = save_dir / 'evolve.csv', save_dir / 'results.csv', save_dir / 'hyp_evolve.yaml'
+ keys = ('metrics/precision', 'metrics/recall', 'metrics/HBBmAP.5', 'metrics/HBBmAP.5:.95',
+ 'val/box_loss', 'val/obj_loss', 'val/cls_loss', 'val/theta_loss') + tuple(hyp.keys()) # [results + hyps]
+ keys = tuple(x.strip() for x in keys)
+ vals = results + tuple(hyp.values())
+ n = len(keys)
+
+ # Download (optional)
+ if bucket:
+ url = f'gs://{bucket}/evolve.csv'
+ if gsutil_getsize(url) > (os.path.getsize(evolve_csv) if os.path.exists(evolve_csv) else 0):
+ os.system(f'gsutil cp {url} {save_dir}') # download evolve.csv if larger than local
+
+ # Log to evolve.csv
+ s = '' if evolve_csv.exists() else (('%20s,' * n % keys).rstrip(',') + '\n') # add header
+ with open(evolve_csv, 'a') as f:
+ f.write(s + ('%20.5g,' * n % vals).rstrip(',') + '\n')
+
+ # Print to screen
+ print(colorstr('evolve: ') + ', '.join(f'{x.strip():>20s}' for x in keys))
+ print(colorstr('evolve: ') + ', '.join(f'{x:20.5g}' for x in vals), end='\n\n\n')
+
+ # Save yaml
+ with open(evolve_yaml, 'w') as f:
+ data = pd.read_csv(evolve_csv)
+ data = data.rename(columns=lambda x: x.strip()) # strip keys
+ i = np.argmax(fitness(data.values[:, :7])) #
+ f.write('# YOLOv5 Hyperparameter Evolution Results\n' +
+ f'# Best generation: {i}\n' +
+ f'# Last generation: {len(data) - 1}\n' +
+ '# ' + ', '.join(f'{x.strip():>20s}' for x in keys[:7]) + '\n' +
+ '# ' + ', '.join(f'{x:>20.5g}' for x in data.values[i, :7]) + '\n\n')
+ yaml.safe_dump(hyp, f, sort_keys=False)
+
+ if bucket:
+ os.system(f'gsutil cp {evolve_csv} {evolve_yaml} gs://{bucket}') # upload
+
+
+def apply_classifier(x, model, img, im0):
+ # Apply a second stage classifier to YOLO outputs
+ # Example model = torchvision.models.__dict__['efficientnet_b0'](pretrained=True).to(device).eval()
+ im0 = [im0] if isinstance(im0, np.ndarray) else im0
+ for i, d in enumerate(x): # per image
+ if d is not None and len(d):
+ d = d.clone()
+
+ # Reshape and pad cutouts
+ b = xyxy2xywh(d[:, :4]) # boxes
+ b[:, 2:] = b[:, 2:].max(1)[0].unsqueeze(1) # rectangle to square
+ b[:, 2:] = b[:, 2:] * 1.3 + 30 # pad
+ d[:, :4] = xywh2xyxy(b).long()
+
+ # Rescale boxes from img_size to im0 size
+ scale_coords(img.shape[2:], d[:, :4], im0[i].shape)
+
+ # Classes
+ pred_cls1 = d[:, 5].long()
+ ims = []
+ for j, a in enumerate(d): # per item
+ cutout = im0[i][int(a[1]):int(a[3]), int(a[0]):int(a[2])]
+ im = cv2.resize(cutout, (224, 224)) # BGR
+ # cv2.imwrite('example%i.jpg' % j, cutout)
+
+ im = im[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
+ im = np.ascontiguousarray(im, dtype=np.float32) # uint8 to float32
+ im /= 255 # 0 - 255 to 0.0 - 1.0
+ ims.append(im)
+
+ pred_cls2 = model(torch.Tensor(ims).to(d.device)).argmax(1) # classifier prediction
+ x[i] = x[i][pred_cls1 == pred_cls2] # retain matching class detections
+
+ return x
+
+
+def increment_path(path, exist_ok=False, sep='', mkdir=False):
+ # Increment file or directory path, i.e. runs/exp --> runs/exp{sep}2, runs/exp{sep}3, ... etc.
+ path = Path(path) # os-agnostic
+ if path.exists() and not exist_ok:
+ path, suffix = (path.with_suffix(''), path.suffix) if path.is_file() else (path, '')
+ dirs = glob.glob(f"{path}{sep}*") # similar paths
+ matches = [re.search(rf"%s{sep}(\d+)" % path.stem, d) for d in dirs]
+ i = [int(m.groups()[0]) for m in matches if m] # indices
+ n = max(i) + 1 if i else 2 # increment number
+ path = Path(f"{path}{sep}{n}{suffix}") # increment path
+ if mkdir:
+ path.mkdir(parents=True, exist_ok=True) # make directory
+ return path
+
+
+# Variables
+NCOLS = 0 if is_docker() else shutil.get_terminal_size().columns # terminal window size for tqdm
diff --git a/utils/google_app_engine/Dockerfile b/utils/google_app_engine/Dockerfile
new file mode 100644
index 0000000..0155618
--- /dev/null
+++ b/utils/google_app_engine/Dockerfile
@@ -0,0 +1,25 @@
+FROM gcr.io/google-appengine/python
+
+# Create a virtualenv for dependencies. This isolates these packages from
+# system-level packages.
+# Use -p python3 or -p python3.7 to select python version. Default is version 2.
+RUN virtualenv /env -p python3
+
+# Setting these environment variables are the same as running
+# source /env/bin/activate.
+ENV VIRTUAL_ENV /env
+ENV PATH /env/bin:$PATH
+
+RUN apt-get update && apt-get install -y python-opencv
+
+# Copy the application's requirements.txt and run pip to install all
+# dependencies into the virtualenv.
+ADD requirements.txt /app/requirements.txt
+RUN pip install -r /app/requirements.txt
+
+# Add the application source code.
+ADD . /app
+
+# Run a WSGI server to serve the application. gunicorn must be declared as
+# a dependency in requirements.txt.
+CMD gunicorn -b :$PORT main:app
diff --git a/utils/google_app_engine/additional_requirements.txt b/utils/google_app_engine/additional_requirements.txt
new file mode 100644
index 0000000..42d7ffc
--- /dev/null
+++ b/utils/google_app_engine/additional_requirements.txt
@@ -0,0 +1,4 @@
+# add these requirements in your app on top of the existing ones
+pip==21.1
+Flask==1.0.2
+gunicorn==19.9.0
diff --git a/utils/google_app_engine/app.yaml b/utils/google_app_engine/app.yaml
new file mode 100644
index 0000000..5056b7c
--- /dev/null
+++ b/utils/google_app_engine/app.yaml
@@ -0,0 +1,14 @@
+runtime: custom
+env: flex
+
+service: yolov5app
+
+liveness_check:
+ initial_delay_sec: 600
+
+manual_scaling:
+ instances: 1
+resources:
+ cpu: 1
+ memory_gb: 4
+ disk_size_gb: 20
diff --git a/utils/loggers/__init__.py b/utils/loggers/__init__.py
new file mode 100644
index 0000000..5186f90
--- /dev/null
+++ b/utils/loggers/__init__.py
@@ -0,0 +1,175 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Logging utils
+"""
+
+import os
+import warnings
+from threading import Thread
+
+import pkg_resources as pkg
+import torch
+from torch.utils.tensorboard import SummaryWriter
+
+from utils.general import colorstr, emojis
+from utils.loggers.wandb.wandb_utils import WandbLogger
+from utils.plots import plot_images, plot_results
+from utils.torch_utils import de_parallel
+
+LOGGERS = ('csv', 'tb', 'wandb') # text-file, TensorBoard, Weights & Biases
+RANK = int(os.getenv('RANK', -1))
+
+try:
+ import wandb
+
+ assert hasattr(wandb, '__version__') # verify package import not local dir
+ if pkg.parse_version(wandb.__version__) >= pkg.parse_version('0.12.2') and RANK in [0, -1]:
+ try:
+ wandb_login_success = wandb.login(timeout=30)
+ except wandb.errors.UsageError: # known non-TTY terminal issue
+ wandb_login_success = False
+ if not wandb_login_success:
+ wandb = None
+except (ImportError, AssertionError):
+ wandb = None
+
+
+class Loggers():
+ # YOLOv5 Loggers class
+ def __init__(self, save_dir=None, weights=None, opt=None, hyp=None, logger=None, include=LOGGERS):
+ self.save_dir = save_dir
+ self.weights = weights
+ self.opt = opt
+ self.hyp = hyp
+ self.logger = logger # for printing results to console
+ self.include = include
+ # self.keys = ['train/box_loss', 'train/obj_loss', 'train/cls_loss', # train loss
+ # 'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95', # metrics
+ # 'val/box_loss', 'val/obj_loss', 'val/cls_loss', # val loss
+ # 'x/lr0', 'x/lr1', 'x/lr2'] # params
+ self.keys = ['train/box_loss', 'train/obj_loss', 'train/cls_loss', 'train/theta_loss', # train loss
+ 'metrics/precision', 'metrics/recall', 'metrics/HBBmAP.5', 'metrics/HBBmAP.5:.95', # metrics
+ 'val/box_loss', 'val/obj_loss', 'val/cls_loss', 'val/theta_loss', # val loss
+ 'x/lr0', 'x/lr1', 'x/lr2'] # params
+ for k in LOGGERS:
+ setattr(self, k, None) # init empty logger dictionary
+ self.csv = True # always log to csv
+
+ # Message
+ if not wandb:
+ prefix = colorstr('Weights & Biases: ')
+ s = f"{prefix}run 'pip install wandb' to automatically track and visualize YOLOv5 🚀 runs (RECOMMENDED)"
+ print(emojis(s))
+
+ # TensorBoard
+ s = self.save_dir
+ if 'tb' in self.include and not self.opt.evolve:
+ prefix = colorstr('TensorBoard: ')
+ self.logger.info(f"{prefix}Start with 'tensorboard --logdir {s.parent}', view at http://localhost:6006/")
+ self.tb = SummaryWriter(str(s))
+
+ # W&B
+ if wandb and 'wandb' in self.include:
+ wandb_artifact_resume = isinstance(self.opt.resume, str) and self.opt.resume.startswith('wandb-artifact://')
+ run_id = torch.load(self.weights).get('wandb_id') if self.opt.resume and not wandb_artifact_resume else None
+ self.opt.hyp = self.hyp # add hyperparameters
+ self.wandb = WandbLogger(self.opt, run_id)
+ else:
+ self.wandb = None
+
+ def on_pretrain_routine_end(self):
+ # Callback runs on pre-train routine end
+ paths = self.save_dir.glob('*labels*.jpg') # training labels
+ if self.wandb:
+ self.wandb.log({"Labels": [wandb.Image(str(x), caption=x.name) for x in paths]})
+
+ def on_train_batch_end(self, ni, model, imgs, targets, paths, plots, sync_bn):
+ """
+ Args:
+ imgs (tensor): (b, 3, height, width)
+ targets (tensor): (n_targets, [img_index clsid cx cy l s theta gaussian_θ_labels])
+ paths (list[str,...]): (b)
+ """
+ # Callback runs on train batch end
+ if plots:
+ if ni == 0:
+ if not sync_bn: # tb.add_graph() --sync known issue https://github.com/ultralytics/yolov5/issues/3754
+ with warnings.catch_warnings():
+ warnings.simplefilter('ignore') # suppress jit trace warning
+ self.tb.add_graph(torch.jit.trace(de_parallel(model), imgs[0:1], strict=False), [])
+ if ni < 8:
+ f = self.save_dir / f'train_batch{ni}.jpg' # filename
+ Thread(target=plot_images, args=(imgs, targets, paths, f), daemon=True).start()
+ if self.wandb and ni == 10:
+ files = sorted(self.save_dir.glob('train*.jpg'))
+ self.wandb.log({'Mosaics': [wandb.Image(str(f), caption=f.name) for f in files if f.exists()]})
+
+ def on_train_epoch_end(self, epoch):
+ # Callback runs on train epoch end
+ if self.wandb:
+ self.wandb.current_epoch = epoch + 1
+
+ def on_val_image_end(self, pred, predn, path, names, im):
+ # Callback runs on val image end
+ if self.wandb:
+ self.wandb.val_one_image(pred, predn, path, names, im)
+
+ def on_val_end(self):
+ # Callback runs on val end
+ if self.wandb:
+ files = sorted(self.save_dir.glob('val*.jpg'))
+ self.wandb.log({"Validation": [wandb.Image(str(f), caption=f.name) for f in files]})
+
+ def on_fit_epoch_end(self, vals, epoch, best_fitness, fi):
+ # Callback runs at the end of each fit (train+val) epoch
+ x = {k: v for k, v in zip(self.keys, vals)} # dict
+ if self.csv:
+ file = self.save_dir / 'results.csv'
+ n = len(x) + 1 # number of cols
+ s = '' if file.exists() else (('%20s,' * n % tuple(['epoch'] + self.keys)).rstrip(',') + '\n') # add header
+ with open(file, 'a') as f:
+ f.write(s + ('%20.5g,' * n % tuple([epoch] + vals)).rstrip(',') + '\n')
+
+ if self.tb:
+ for k, v in x.items():
+ self.tb.add_scalar(k, v, epoch)
+
+ if self.wandb:
+ self.wandb.log(x)
+ self.wandb.end_epoch(best_result=best_fitness == fi)
+
+ def on_model_save(self, last, epoch, final_epoch, best_fitness, fi):
+ # Callback runs on model save event
+ if self.wandb:
+ if ((epoch + 1) % self.opt.save_period == 0 and not final_epoch) and self.opt.save_period != -1:
+ self.wandb.log_model(last.parent, self.opt, epoch, fi, best_model=best_fitness == fi)
+
+ def on_train_end(self, last, best, plots, epoch, results):
+ # Callback runs on training end
+ if plots:
+ plot_results(file=self.save_dir / 'results.csv') # save results.png
+ files = ['results.png', 'confusion_matrix.png', *(f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R'))]
+ files = [(self.save_dir / f) for f in files if (self.save_dir / f).exists()] # filter
+
+ if self.tb:
+ import cv2
+ for f in files:
+ self.tb.add_image(f.stem, cv2.imread(str(f))[..., ::-1], epoch, dataformats='HWC')
+
+ if self.wandb:
+ self.wandb.log({"Results": [wandb.Image(str(f), caption=f.name) for f in files]})
+ # Calling wandb.log. TODO: Refactor this into WandbLogger.log_model
+ if not self.opt.evolve:
+ wandb.log_artifact(str(best if best.exists() else last), type='model',
+ name='run_' + self.wandb.wandb_run.id + '_model',
+ aliases=['latest', 'best', 'stripped'])
+ self.wandb.finish_run()
+ else:
+ self.wandb.finish_run()
+ self.wandb = WandbLogger(self.opt)
+
+ def on_params_update(self, params):
+ # Update hyperparams or configs of the experiment
+ # params: A dict containing {param: value} pairs
+ if self.wandb:
+ self.wandb.wandb_run.config.update(params, allow_val_change=True)
diff --git a/utils/loggers/__pycache__/__init__.cpython-37.pyc b/utils/loggers/__pycache__/__init__.cpython-37.pyc
new file mode 100644
index 0000000..9021f3f
Binary files /dev/null and b/utils/loggers/__pycache__/__init__.cpython-37.pyc differ
diff --git a/utils/loggers/wandb/README.md b/utils/loggers/wandb/README.md
new file mode 100644
index 0000000..63d9998
--- /dev/null
+++ b/utils/loggers/wandb/README.md
@@ -0,0 +1,152 @@
+📚 This guide explains how to use **Weights & Biases** (W&B) with YOLOv5 🚀. UPDATED 29 September 2021.
+* [About Weights & Biases](#about-weights-&-biases)
+* [First-Time Setup](#first-time-setup)
+* [Viewing runs](#viewing-runs)
+* [Disabling wandb](#disabling-wandb)
+* [Advanced Usage: Dataset Versioning and Evaluation](#advanced-usage)
+* [Reports: Share your work with the world!](#reports)
+
+## About Weights & Biases
+Think of [W&B](https://wandb.ai/site?utm_campaign=repo_yolo_wandbtutorial) like GitHub for machine learning models. With a few lines of code, save everything you need to debug, compare and reproduce your models — architecture, hyperparameters, git commits, model weights, GPU usage, and even datasets and predictions.
+
+Used by top researchers including teams at OpenAI, Lyft, Github, and MILA, W&B is part of the new standard of best practices for machine learning. How W&B can help you optimize your machine learning workflows:
+
+ * [Debug](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#Free-2) model performance in real time
+ * [GPU usage](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#System-4) visualized automatically
+ * [Custom charts](https://wandb.ai/wandb/customizable-charts/reports/Powerful-Custom-Charts-To-Debug-Model-Peformance--VmlldzoyNzY4ODI) for powerful, extensible visualization
+ * [Share insights](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#Share-8) interactively with collaborators
+ * [Optimize hyperparameters](https://docs.wandb.com/sweeps) efficiently
+ * [Track](https://docs.wandb.com/artifacts) datasets, pipelines, and production models
+
+## First-Time Setup
+
+ Toggle Details
+When you first train, W&B will prompt you to create a new account and will generate an **API key** for you. If you are an existing user you can retrieve your key from https://wandb.ai/authorize. This key is used to tell W&B where to log your data. You only need to supply your key once, and then it is remembered on the same device.
+
+W&B will create a cloud **project** (default is 'YOLOv5') for your training runs, and each new training run will be provided a unique run **name** within that project as project/name. You can also manually set your project and run name as:
+
+ ```shell
+ $ python train.py --project ... --name ...
+ ```
+
+YOLOv5 notebook example:
+
+
+
+
+
+## Viewing Runs
+
+ Toggle Details
+Run information streams from your environment to the W&B cloud console as you train. This allows you to monitor and even cancel runs in realtime . All important information is logged:
+
+ * Training & Validation losses
+ * Metrics: Precision, Recall, mAP@0.5, mAP@0.5:0.95
+ * Learning Rate over time
+ * A bounding box debugging panel, showing the training progress over time
+ * GPU: Type, **GPU Utilization**, power, temperature, **CUDA memory usage**
+ * System: Disk I/0, CPU utilization, RAM memory usage
+ * Your trained model as W&B Artifact
+ * Environment: OS and Python types, Git repository and state, **training command**
+
+
+
+
+ ## Disabling wandb
+* training after running `wandb disabled` inside that directory creates no wandb run
+![Screenshot (84)](https://user-images.githubusercontent.com/15766192/143441777-c780bdd7-7cb4-4404-9559-b4316030a985.png)
+
+* To enable wandb again, run `wandb online`
+![Screenshot (85)](https://user-images.githubusercontent.com/15766192/143441866-7191b2cb-22f0-4e0f-ae64-2dc47dc13078.png)
+
+## Advanced Usage
+You can leverage W&B artifacts and Tables integration to easily visualize and manage your datasets, models and training evaluations. Here are some quick examples to get you started.
+
+
1: Train and Log Evaluation simultaneousy
+ This is an extension of the previous section, but it'll also training after uploading the dataset. This also evaluation Table
+ Evaluation table compares your predictions and ground truths across the validation set for each epoch. It uses the references to the already uploaded datasets,
+ so no images will be uploaded from your system more than once.
+
+ Usage
+ Code $ python train.py --upload_data val
+
+![Screenshot from 2021-11-21 17-40-06](https://user-images.githubusercontent.com/15766192/142761183-c1696d8c-3f38-45ab-991a-bb0dfd98ae7d.png)
+
+
+
2. Visualize and Version Datasets
+ Log, visualize, dynamically query, and understand your data with W&B Tables. You can use the following command to log your dataset as a W&B Table. This will generate a {dataset}_wandb.yaml file which can be used to train from dataset artifact.
+
+ Usage
+ Code $ python utils/logger/wandb/log_dataset.py --project ... --name ... --data ..
+
+ ![Screenshot (64)](https://user-images.githubusercontent.com/15766192/128486078-d8433890-98a3-4d12-8986-b6c0e3fc64b9.png)
+
+
+
3: Train using dataset artifact
+ When you upload a dataset as described in the first section, you get a new config file with an added `_wandb` to its name. This file contains the information that
+ can be used to train a model directly from the dataset artifact. This also logs evaluation
+
+ Usage
+ Code $ python train.py --data {data}_wandb.yaml
+
+![Screenshot (72)](https://user-images.githubusercontent.com/15766192/128979739-4cf63aeb-a76f-483f-8861-1c0100b938a5.png)
+
+
+
4: Save model checkpoints as artifacts
+ To enable saving and versioning checkpoints of your experiment, pass `--save_period n` with the base cammand, where `n` represents checkpoint interval.
+ You can also log both the dataset and model checkpoints simultaneously. If not passed, only the final model will be logged
+
+
+ Usage
+ Code $ python train.py --save_period 1
+
+![Screenshot (68)](https://user-images.githubusercontent.com/15766192/128726138-ec6c1f60-639d-437d-b4ee-3acd9de47ef3.png)
+
+
+
+
+
5: Resume runs from checkpoint artifacts.
+Any run can be resumed using artifacts if the --resume argument starts with wandb-artifact:// prefix followed by the run path, i.e, wandb-artifact://username/project/runid . This doesn't require the model checkpoint to be present on the local system.
+
+
+ Usage
+ Code $ python train.py --resume wandb-artifact://{run_path}
+
+![Screenshot (70)](https://user-images.githubusercontent.com/15766192/128728988-4e84b355-6c87-41ae-a591-14aecf45343e.png)
+
+
+
6: Resume runs from dataset artifact & checkpoint artifacts.
+ Local dataset or model checkpoints are not required. This can be used to resume runs directly on a different device
+ The syntax is same as the previous section, but you'll need to lof both the dataset and model checkpoints as artifacts, i.e, set bot --upload_dataset or
+ train from _wandb.yaml file and set --save_period
+
+
+ Usage
+ Code $ python train.py --resume wandb-artifact://{run_path}
+
+![Screenshot (70)](https://user-images.githubusercontent.com/15766192/128728988-4e84b355-6c87-41ae-a591-14aecf45343e.png)
+
+
+
+
+
Reports
+W&B Reports can be created from your saved runs for sharing online. Once a report is created you will receive a link you can use to publically share your results. Here is an example report created from the COCO128 tutorial trainings of all four YOLOv5 models ([link](https://wandb.ai/glenn-jocher/yolov5_tutorial/reports/YOLOv5-COCO128-Tutorial-Results--VmlldzozMDI5OTY)).
+
+
+
+
+## Environments
+
+YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including [CUDA](https://developer.nvidia.com/cuda)/[CUDNN](https://developer.nvidia.com/cudnn), [Python](https://www.python.org/) and [PyTorch](https://pytorch.org/) preinstalled):
+
+- **Google Colab and Kaggle** notebooks with free GPU:
+- **Google Cloud** Deep Learning VM. See [GCP Quickstart Guide](https://github.com/ultralytics/yolov5/wiki/GCP-Quickstart)
+- **Amazon** Deep Learning AMI. See [AWS Quickstart Guide](https://github.com/ultralytics/yolov5/wiki/AWS-Quickstart)
+- **Docker Image**. See [Docker Quickstart Guide](https://github.com/ultralytics/yolov5/wiki/Docker-Quickstart)
+
+
+## Status
+
+![CI CPU testing](https://github.com/ultralytics/yolov5/workflows/CI%20CPU%20testing/badge.svg)
+
+If this badge is green, all [YOLOv5 GitHub Actions](https://github.com/ultralytics/yolov5/actions) Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training ([train.py](https://github.com/ultralytics/yolov5/blob/master/train.py)), validation ([val.py](https://github.com/ultralytics/yolov5/blob/master/val.py)), inference ([detect.py](https://github.com/ultralytics/yolov5/blob/master/detect.py)) and export ([export.py](https://github.com/ultralytics/yolov5/blob/master/export.py)) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.
diff --git a/utils/loggers/wandb/__init__.py b/utils/loggers/wandb/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/utils/loggers/wandb/__pycache__/__init__.cpython-37.pyc b/utils/loggers/wandb/__pycache__/__init__.cpython-37.pyc
new file mode 100644
index 0000000..462122a
Binary files /dev/null and b/utils/loggers/wandb/__pycache__/__init__.cpython-37.pyc differ
diff --git a/utils/loggers/wandb/__pycache__/__init__.cpython-39.pyc b/utils/loggers/wandb/__pycache__/__init__.cpython-39.pyc
new file mode 100644
index 0000000..862cb6d
Binary files /dev/null and b/utils/loggers/wandb/__pycache__/__init__.cpython-39.pyc differ
diff --git a/utils/loggers/wandb/__pycache__/wandb_utils.cpython-37.pyc b/utils/loggers/wandb/__pycache__/wandb_utils.cpython-37.pyc
new file mode 100644
index 0000000..5988ad1
Binary files /dev/null and b/utils/loggers/wandb/__pycache__/wandb_utils.cpython-37.pyc differ
diff --git a/utils/loggers/wandb/__pycache__/wandb_utils.cpython-39.pyc b/utils/loggers/wandb/__pycache__/wandb_utils.cpython-39.pyc
new file mode 100644
index 0000000..f58a4be
Binary files /dev/null and b/utils/loggers/wandb/__pycache__/wandb_utils.cpython-39.pyc differ
diff --git a/utils/loggers/wandb/log_dataset.py b/utils/loggers/wandb/log_dataset.py
new file mode 100644
index 0000000..06e81fb
--- /dev/null
+++ b/utils/loggers/wandb/log_dataset.py
@@ -0,0 +1,27 @@
+import argparse
+
+from wandb_utils import WandbLogger
+
+from utils.general import LOGGER
+
+WANDB_ARTIFACT_PREFIX = 'wandb-artifact://'
+
+
+def create_dataset_artifact(opt):
+ logger = WandbLogger(opt, None, job_type='Dataset Creation') # TODO: return value unused
+ if not logger.wandb:
+ LOGGER.info("install wandb using `pip install wandb` to log the dataset")
+
+
+if __name__ == '__main__':
+ parser = argparse.ArgumentParser()
+ parser.add_argument('--data', type=str, default='data/coco128.yaml', help='data.yaml path')
+ parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
+ parser.add_argument('--project', type=str, default='YOLOv5', help='name of W&B Project')
+ parser.add_argument('--entity', default=None, help='W&B entity')
+ parser.add_argument('--name', type=str, default='log dataset', help='name of W&B run')
+
+ opt = parser.parse_args()
+ opt.resume = False # Explicitly disallow resume check for dataset upload job
+
+ create_dataset_artifact(opt)
diff --git a/utils/loggers/wandb/sweep.py b/utils/loggers/wandb/sweep.py
new file mode 100644
index 0000000..206059b
--- /dev/null
+++ b/utils/loggers/wandb/sweep.py
@@ -0,0 +1,41 @@
+import sys
+from pathlib import Path
+
+import wandb
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[3] # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+ sys.path.append(str(ROOT)) # add ROOT to PATH
+
+from train import parse_opt, train
+from utils.callbacks import Callbacks
+from utils.general import increment_path
+from utils.torch_utils import select_device
+
+
+def sweep():
+ wandb.init()
+ # Get hyp dict from sweep agent
+ hyp_dict = vars(wandb.config).get("_items")
+
+ # Workaround: get necessary opt args
+ opt = parse_opt(known=True)
+ opt.batch_size = hyp_dict.get("batch_size")
+ opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok or opt.evolve))
+ opt.epochs = hyp_dict.get("epochs")
+ opt.nosave = True
+ opt.data = hyp_dict.get("data")
+ opt.weights = str(opt.weights)
+ opt.cfg = str(opt.cfg)
+ opt.data = str(opt.data)
+ opt.hyp = str(opt.hyp)
+ opt.project = str(opt.project)
+ device = select_device(opt.device, batch_size=opt.batch_size)
+
+ # train
+ train(hyp_dict, opt, device, callbacks=Callbacks())
+
+
+if __name__ == "__main__":
+ sweep()
diff --git a/utils/loggers/wandb/sweep.yaml b/utils/loggers/wandb/sweep.yaml
new file mode 100644
index 0000000..c7790d7
--- /dev/null
+++ b/utils/loggers/wandb/sweep.yaml
@@ -0,0 +1,143 @@
+# Hyperparameters for training
+# To set range-
+# Provide min and max values as:
+# parameter:
+#
+# min: scalar
+# max: scalar
+# OR
+#
+# Set a specific list of search space-
+# parameter:
+# values: [scalar1, scalar2, scalar3...]
+#
+# You can use grid, bayesian and hyperopt search strategy
+# For more info on configuring sweeps visit - https://docs.wandb.ai/guides/sweeps/configuration
+
+program: utils/loggers/wandb/sweep.py
+method: random
+metric:
+ name: metrics/mAP_0.5
+ goal: maximize
+
+parameters:
+ # hyperparameters: set either min, max range or values list
+ data:
+ value: "data/coco128.yaml"
+ batch_size:
+ values: [64]
+ epochs:
+ values: [10]
+
+ lr0:
+ distribution: uniform
+ min: 1e-5
+ max: 1e-1
+ lrf:
+ distribution: uniform
+ min: 0.01
+ max: 1.0
+ momentum:
+ distribution: uniform
+ min: 0.6
+ max: 0.98
+ weight_decay:
+ distribution: uniform
+ min: 0.0
+ max: 0.001
+ warmup_epochs:
+ distribution: uniform
+ min: 0.0
+ max: 5.0
+ warmup_momentum:
+ distribution: uniform
+ min: 0.0
+ max: 0.95
+ warmup_bias_lr:
+ distribution: uniform
+ min: 0.0
+ max: 0.2
+ box:
+ distribution: uniform
+ min: 0.02
+ max: 0.2
+ cls:
+ distribution: uniform
+ min: 0.2
+ max: 4.0
+ cls_pw:
+ distribution: uniform
+ min: 0.5
+ max: 2.0
+ obj:
+ distribution: uniform
+ min: 0.2
+ max: 4.0
+ obj_pw:
+ distribution: uniform
+ min: 0.5
+ max: 2.0
+ iou_t:
+ distribution: uniform
+ min: 0.1
+ max: 0.7
+ anchor_t:
+ distribution: uniform
+ min: 2.0
+ max: 8.0
+ fl_gamma:
+ distribution: uniform
+ min: 0.0
+ max: 0.1
+ hsv_h:
+ distribution: uniform
+ min: 0.0
+ max: 0.1
+ hsv_s:
+ distribution: uniform
+ min: 0.0
+ max: 0.9
+ hsv_v:
+ distribution: uniform
+ min: 0.0
+ max: 0.9
+ degrees:
+ distribution: uniform
+ min: 0.0
+ max: 45.0
+ translate:
+ distribution: uniform
+ min: 0.0
+ max: 0.9
+ scale:
+ distribution: uniform
+ min: 0.0
+ max: 0.9
+ shear:
+ distribution: uniform
+ min: 0.0
+ max: 10.0
+ perspective:
+ distribution: uniform
+ min: 0.0
+ max: 0.001
+ flipud:
+ distribution: uniform
+ min: 0.0
+ max: 1.0
+ fliplr:
+ distribution: uniform
+ min: 0.0
+ max: 1.0
+ mosaic:
+ distribution: uniform
+ min: 0.0
+ max: 1.0
+ mixup:
+ distribution: uniform
+ min: 0.0
+ max: 1.0
+ copy_paste:
+ distribution: uniform
+ min: 0.0
+ max: 1.0
diff --git a/utils/loggers/wandb/wandb_utils.py b/utils/loggers/wandb/wandb_utils.py
new file mode 100644
index 0000000..221d3c8
--- /dev/null
+++ b/utils/loggers/wandb/wandb_utils.py
@@ -0,0 +1,560 @@
+"""Utilities and tools for tracking runs with Weights & Biases."""
+
+import logging
+import os
+import sys
+from contextlib import contextmanager
+from pathlib import Path
+from typing import Dict
+
+import yaml
+from tqdm import tqdm
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[3] # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+ sys.path.append(str(ROOT)) # add ROOT to PATH
+
+from utils.datasets import LoadImagesAndLabels, img2label_paths
+from utils.general import LOGGER, check_dataset, check_file
+
+try:
+ import wandb
+
+ assert hasattr(wandb, '__version__') # verify package import not local dir
+except (ImportError, AssertionError):
+ wandb = None
+
+RANK = int(os.getenv('RANK', -1))
+WANDB_ARTIFACT_PREFIX = 'wandb-artifact://'
+
+
+def remove_prefix(from_string, prefix=WANDB_ARTIFACT_PREFIX):
+ return from_string[len(prefix):]
+
+
+def check_wandb_config_file(data_config_file):
+ wandb_config = '_wandb.'.join(data_config_file.rsplit('.', 1)) # updated data.yaml path
+ if Path(wandb_config).is_file():
+ return wandb_config
+ return data_config_file
+
+
+def check_wandb_dataset(data_file):
+ is_trainset_wandb_artifact = False
+ is_valset_wandb_artifact = False
+ if check_file(data_file) and data_file.endswith('.yaml'):
+ with open(data_file, errors='ignore') as f:
+ data_dict = yaml.safe_load(f)
+ is_trainset_wandb_artifact = (isinstance(data_dict['train'], str) and
+ data_dict['train'].startswith(WANDB_ARTIFACT_PREFIX))
+ is_valset_wandb_artifact = (isinstance(data_dict['val'], str) and
+ data_dict['val'].startswith(WANDB_ARTIFACT_PREFIX))
+ if is_trainset_wandb_artifact or is_valset_wandb_artifact:
+ return data_dict
+ else:
+ return check_dataset(data_file)
+
+
+def get_run_info(run_path):
+ run_path = Path(remove_prefix(run_path, WANDB_ARTIFACT_PREFIX))
+ run_id = run_path.stem
+ project = run_path.parent.stem
+ entity = run_path.parent.parent.stem
+ model_artifact_name = 'run_' + run_id + '_model'
+ return entity, project, run_id, model_artifact_name
+
+
+def check_wandb_resume(opt):
+ process_wandb_config_ddp_mode(opt) if RANK not in [-1, 0] else None
+ if isinstance(opt.resume, str):
+ if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
+ if RANK not in [-1, 0]: # For resuming DDP runs
+ entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
+ api = wandb.Api()
+ artifact = api.artifact(entity + '/' + project + '/' + model_artifact_name + ':latest')
+ modeldir = artifact.download()
+ opt.weights = str(Path(modeldir) / "last.pt")
+ return True
+ return None
+
+
+def process_wandb_config_ddp_mode(opt):
+ with open(check_file(opt.data), errors='ignore') as f:
+ data_dict = yaml.safe_load(f) # data dict
+ train_dir, val_dir = None, None
+ if isinstance(data_dict['train'], str) and data_dict['train'].startswith(WANDB_ARTIFACT_PREFIX):
+ api = wandb.Api()
+ train_artifact = api.artifact(remove_prefix(data_dict['train']) + ':' + opt.artifact_alias)
+ train_dir = train_artifact.download()
+ train_path = Path(train_dir) / 'data/images/'
+ data_dict['train'] = str(train_path)
+
+ if isinstance(data_dict['val'], str) and data_dict['val'].startswith(WANDB_ARTIFACT_PREFIX):
+ api = wandb.Api()
+ val_artifact = api.artifact(remove_prefix(data_dict['val']) + ':' + opt.artifact_alias)
+ val_dir = val_artifact.download()
+ val_path = Path(val_dir) / 'data/images/'
+ data_dict['val'] = str(val_path)
+ if train_dir or val_dir:
+ ddp_data_path = str(Path(val_dir) / 'wandb_local_data.yaml')
+ with open(ddp_data_path, 'w') as f:
+ yaml.safe_dump(data_dict, f)
+ opt.data = ddp_data_path
+
+
+class WandbLogger():
+ """Log training runs, datasets, models, and predictions to Weights & Biases.
+
+ This logger sends information to W&B at wandb.ai. By default, this information
+ includes hyperparameters, system configuration and metrics, model metrics,
+ and basic data metrics and analyses.
+
+ By providing additional command line arguments to train.py, datasets,
+ models and predictions can also be logged.
+
+ For more on how this logger is used, see the Weights & Biases documentation:
+ https://docs.wandb.com/guides/integrations/yolov5
+ """
+
+ def __init__(self, opt, run_id=None, job_type='Training'):
+ """
+ - Initialize WandbLogger instance
+ - Upload dataset if opt.upload_dataset is True
+ - Setup trainig processes if job_type is 'Training'
+
+ arguments:
+ opt (namespace) -- Commandline arguments for this run
+ run_id (str) -- Run ID of W&B run to be resumed
+ job_type (str) -- To set the job_type for this run
+
+ """
+ # Pre-training routine --
+ self.job_type = job_type
+ self.wandb, self.wandb_run = wandb, None if not wandb else wandb.run
+ self.val_artifact, self.train_artifact = None, None
+ self.train_artifact_path, self.val_artifact_path = None, None
+ self.result_artifact = None
+ self.val_table, self.result_table = None, None
+ self.bbox_media_panel_images = []
+ self.val_table_path_map = None
+ self.max_imgs_to_log = 16
+ self.wandb_artifact_data_dict = None
+ self.data_dict = None
+ # It's more elegant to stick to 1 wandb.init call,
+ # but useful config data is overwritten in the WandbLogger's wandb.init call
+ if isinstance(opt.resume, str): # checks resume from artifact
+ if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
+ entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
+ model_artifact_name = WANDB_ARTIFACT_PREFIX + model_artifact_name
+ assert wandb, 'install wandb to resume wandb runs'
+ # Resume wandb-artifact:// runs here| workaround for not overwriting wandb.config
+ self.wandb_run = wandb.init(id=run_id,
+ project=project,
+ entity=entity,
+ resume='allow',
+ allow_val_change=True)
+ opt.resume = model_artifact_name
+ elif self.wandb:
+ self.wandb_run = wandb.init(config=opt,
+ resume="allow",
+ project='YOLOv5' if opt.project == 'runs/train' else Path(opt.project).stem,
+ entity=opt.entity,
+ name=opt.name if opt.name != 'exp' else None,
+ job_type=job_type,
+ id=run_id,
+ allow_val_change=True) if not wandb.run else wandb.run
+ if self.wandb_run:
+ if self.job_type == 'Training':
+ if opt.upload_dataset:
+ if not opt.resume:
+ self.wandb_artifact_data_dict = self.check_and_upload_dataset(opt)
+
+ if opt.resume:
+ # resume from artifact
+ if isinstance(opt.resume, str) and opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
+ self.data_dict = dict(self.wandb_run.config.data_dict)
+ else: # local resume
+ self.data_dict = check_wandb_dataset(opt.data)
+ else:
+ self.data_dict = check_wandb_dataset(opt.data)
+ self.wandb_artifact_data_dict = self.wandb_artifact_data_dict or self.data_dict
+
+ # write data_dict to config. useful for resuming from artifacts. Do this only when not resuming.
+ self.wandb_run.config.update({'data_dict': self.wandb_artifact_data_dict},
+ allow_val_change=True)
+ self.setup_training(opt)
+
+ if self.job_type == 'Dataset Creation':
+ self.wandb_run.config.update({"upload_dataset": True})
+ self.data_dict = self.check_and_upload_dataset(opt)
+
+ def check_and_upload_dataset(self, opt):
+ """
+ Check if the dataset format is compatible and upload it as W&B artifact
+
+ arguments:
+ opt (namespace)-- Commandline arguments for current run
+
+ returns:
+ Updated dataset info dictionary where local dataset paths are replaced by WAND_ARFACT_PREFIX links.
+ """
+ assert wandb, 'Install wandb to upload dataset'
+ config_path = self.log_dataset_artifact(opt.data,
+ opt.single_cls,
+ 'YOLOv5' if opt.project == 'runs/train' else Path(opt.project).stem)
+ with open(config_path, errors='ignore') as f:
+ wandb_data_dict = yaml.safe_load(f)
+ return wandb_data_dict
+
+ def setup_training(self, opt):
+ """
+ Setup the necessary processes for training YOLO models:
+ - Attempt to download model checkpoint and dataset artifacts if opt.resume stats with WANDB_ARTIFACT_PREFIX
+ - Update data_dict, to contain info of previous run if resumed and the paths of dataset artifact if downloaded
+ - Setup log_dict, initialize bbox_interval
+
+ arguments:
+ opt (namespace) -- commandline arguments for this run
+
+ """
+ self.log_dict, self.current_epoch = {}, 0
+ self.bbox_interval = opt.bbox_interval
+ if isinstance(opt.resume, str):
+ modeldir, _ = self.download_model_artifact(opt)
+ if modeldir:
+ self.weights = Path(modeldir) / "last.pt"
+ config = self.wandb_run.config
+ opt.weights, opt.save_period, opt.batch_size, opt.bbox_interval, opt.epochs, opt.hyp = str(
+ self.weights), config.save_period, config.batch_size, config.bbox_interval, config.epochs, \
+ config.hyp
+ data_dict = self.data_dict
+ if self.val_artifact is None: # If --upload_dataset is set, use the existing artifact, don't download
+ self.train_artifact_path, self.train_artifact = self.download_dataset_artifact(data_dict.get('train'),
+ opt.artifact_alias)
+ self.val_artifact_path, self.val_artifact = self.download_dataset_artifact(data_dict.get('val'),
+ opt.artifact_alias)
+
+ if self.train_artifact_path is not None:
+ train_path = Path(self.train_artifact_path) / 'data/images/'
+ data_dict['train'] = str(train_path)
+ if self.val_artifact_path is not None:
+ val_path = Path(self.val_artifact_path) / 'data/images/'
+ data_dict['val'] = str(val_path)
+
+ if self.val_artifact is not None:
+ self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
+ columns = ["epoch", "id", "ground truth", "prediction"]
+ columns.extend(self.data_dict['names'])
+ self.result_table = wandb.Table(columns)
+ self.val_table = self.val_artifact.get("val")
+ if self.val_table_path_map is None:
+ self.map_val_table_path()
+ if opt.bbox_interval == -1:
+ self.bbox_interval = opt.bbox_interval = (opt.epochs // 10) if opt.epochs > 10 else 1
+ train_from_artifact = self.train_artifact_path is not None and self.val_artifact_path is not None
+ # Update the the data_dict to point to local artifacts dir
+ if train_from_artifact:
+ self.data_dict = data_dict
+
+ def download_dataset_artifact(self, path, alias):
+ """
+ download the model checkpoint artifact if the path starts with WANDB_ARTIFACT_PREFIX
+
+ arguments:
+ path -- path of the dataset to be used for training
+ alias (str)-- alias of the artifact to be download/used for training
+
+ returns:
+ (str, wandb.Artifact) -- path of the downladed dataset and it's corresponding artifact object if dataset
+ is found otherwise returns (None, None)
+ """
+ if isinstance(path, str) and path.startswith(WANDB_ARTIFACT_PREFIX):
+ artifact_path = Path(remove_prefix(path, WANDB_ARTIFACT_PREFIX) + ":" + alias)
+ dataset_artifact = wandb.use_artifact(artifact_path.as_posix().replace("\\", "/"))
+ assert dataset_artifact is not None, "'Error: W&B dataset artifact doesn\'t exist'"
+ datadir = dataset_artifact.download()
+ return datadir, dataset_artifact
+ return None, None
+
+ def download_model_artifact(self, opt):
+ """
+ download the model checkpoint artifact if the resume path starts with WANDB_ARTIFACT_PREFIX
+
+ arguments:
+ opt (namespace) -- Commandline arguments for this run
+ """
+ if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
+ model_artifact = wandb.use_artifact(remove_prefix(opt.resume, WANDB_ARTIFACT_PREFIX) + ":latest")
+ assert model_artifact is not None, 'Error: W&B model artifact doesn\'t exist'
+ modeldir = model_artifact.download()
+ epochs_trained = model_artifact.metadata.get('epochs_trained')
+ total_epochs = model_artifact.metadata.get('total_epochs')
+ is_finished = total_epochs is None
+ assert not is_finished, 'training is finished, can only resume incomplete runs.'
+ return modeldir, model_artifact
+ return None, None
+
+ def log_model(self, path, opt, epoch, fitness_score, best_model=False):
+ """
+ Log the model checkpoint as W&B artifact
+
+ arguments:
+ path (Path) -- Path of directory containing the checkpoints
+ opt (namespace) -- Command line arguments for this run
+ epoch (int) -- Current epoch number
+ fitness_score (float) -- fitness score for current epoch
+ best_model (boolean) -- Boolean representing if the current checkpoint is the best yet.
+ """
+ model_artifact = wandb.Artifact('run_' + wandb.run.id + '_model', type='model', metadata={
+ 'original_url': str(path),
+ 'epochs_trained': epoch + 1,
+ 'save period': opt.save_period,
+ 'project': opt.project,
+ 'total_epochs': opt.epochs,
+ 'fitness_score': fitness_score
+ })
+ model_artifact.add_file(str(path / 'last.pt'), name='last.pt')
+ wandb.log_artifact(model_artifact,
+ aliases=['latest', 'last', 'epoch ' + str(self.current_epoch), 'best' if best_model else ''])
+ LOGGER.info(f"Saving model artifact on epoch {epoch + 1}")
+
+ def log_dataset_artifact(self, data_file, single_cls, project, overwrite_config=False):
+ """
+ Log the dataset as W&B artifact and return the new data file with W&B links
+
+ arguments:
+ data_file (str) -- the .yaml file with information about the dataset like - path, classes etc.
+ single_class (boolean) -- train multi-class data as single-class
+ project (str) -- project name. Used to construct the artifact path
+ overwrite_config (boolean) -- overwrites the data.yaml file if set to true otherwise creates a new
+ file with _wandb postfix. Eg -> data_wandb.yaml
+
+ returns:
+ the new .yaml file with artifact links. it can be used to start training directly from artifacts
+ """
+ upload_dataset = self.wandb_run.config.upload_dataset
+ log_val_only = isinstance(upload_dataset, str) and upload_dataset == 'val'
+ self.data_dict = check_dataset(data_file) # parse and check
+ data = dict(self.data_dict)
+ nc, names = (1, ['item']) if single_cls else (int(data['nc']), data['names'])
+ names = {k: v for k, v in enumerate(names)} # to index dictionary
+
+ # log train set
+ if not log_val_only:
+ self.train_artifact = self.create_dataset_table(LoadImagesAndLabels(
+ data['train'], rect=True, batch_size=1), names, name='train') if data.get('train') else None
+ if data.get('train'):
+ data['train'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'train')
+
+ self.val_artifact = self.create_dataset_table(LoadImagesAndLabels(
+ data['val'], rect=True, batch_size=1), names, name='val') if data.get('val') else None
+ if data.get('val'):
+ data['val'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'val')
+
+ path = Path(data_file)
+ # create a _wandb.yaml file with artifacts links if both train and test set are logged
+ if not log_val_only:
+ path = (path.stem if overwrite_config else path.stem + '_wandb') + '.yaml' # updated data.yaml path
+ path = Path('data') / path
+ data.pop('download', None)
+ data.pop('path', None)
+ with open(path, 'w') as f:
+ yaml.safe_dump(data, f)
+ LOGGER.info(f"Created dataset config file {path}")
+
+ if self.job_type == 'Training': # builds correct artifact pipeline graph
+ if not log_val_only:
+ self.wandb_run.log_artifact(
+ self.train_artifact) # calling use_artifact downloads the dataset. NOT NEEDED!
+ self.wandb_run.use_artifact(self.val_artifact)
+ self.val_artifact.wait()
+ self.val_table = self.val_artifact.get('val')
+ self.map_val_table_path()
+ else:
+ self.wandb_run.log_artifact(self.train_artifact)
+ self.wandb_run.log_artifact(self.val_artifact)
+ return path
+
+ def map_val_table_path(self):
+ """
+ Map the validation dataset Table like name of file -> it's id in the W&B Table.
+ Useful for - referencing artifacts for evaluation.
+ """
+ self.val_table_path_map = {}
+ LOGGER.info("Mapping dataset")
+ for i, data in enumerate(tqdm(self.val_table.data)):
+ self.val_table_path_map[data[3]] = data[0]
+
+ def create_dataset_table(self, dataset: LoadImagesAndLabels, class_to_id: Dict[int, str], name: str = 'dataset'):
+ """
+ Create and return W&B artifact containing W&B Table of the dataset.
+
+ arguments:
+ dataset -- instance of LoadImagesAndLabels class used to iterate over the data to build Table
+ class_to_id -- hash map that maps class ids to labels
+ name -- name of the artifact
+
+ returns:
+ dataset artifact to be logged or used
+ """
+ # TODO: Explore multiprocessing to slpit this loop parallely| This is essential for speeding up the the logging
+ artifact = wandb.Artifact(name=name, type="dataset")
+ img_files = tqdm([dataset.path]) if isinstance(dataset.path, str) and Path(dataset.path).is_dir() else None
+ img_files = tqdm(dataset.img_files) if not img_files else img_files
+ for img_file in img_files:
+ if Path(img_file).is_dir():
+ artifact.add_dir(img_file, name='data/images')
+ labels_path = 'labels'.join(dataset.path.rsplit('images', 1))
+ artifact.add_dir(labels_path, name='data/labels')
+ else:
+ artifact.add_file(img_file, name='data/images/' + Path(img_file).name)
+ label_file = Path(img2label_paths([img_file])[0])
+ artifact.add_file(str(label_file),
+ name='data/labels/' + label_file.name) if label_file.exists() else None
+ table = wandb.Table(columns=["id", "train_image", "Classes", "name"])
+ class_set = wandb.Classes([{'id': id, 'name': name} for id, name in class_to_id.items()])
+ for si, (img, labels, paths, shapes) in enumerate(tqdm(dataset)):
+ box_data, img_classes = [], {}
+ for cls, *xywh in labels[:, 1:].tolist():
+ cls = int(cls)
+ box_data.append({"position": {"middle": [xywh[0], xywh[1]], "width": xywh[2], "height": xywh[3]},
+ "class_id": cls,
+ "box_caption": "%s" % (class_to_id[cls])})
+ img_classes[cls] = class_to_id[cls]
+ boxes = {"ground_truth": {"box_data": box_data, "class_labels": class_to_id}} # inference-space
+ table.add_data(si, wandb.Image(paths, classes=class_set, boxes=boxes), list(img_classes.values()),
+ Path(paths).name)
+ artifact.add(table, name)
+ return artifact
+
+ def log_training_progress(self, predn, path, names):
+ """
+ Build evaluation Table. Uses reference from validation dataset table.
+
+ arguments:
+ predn (list): list of predictions in the native space in the format - [xmin, ymin, xmax, ymax, confidence, class]
+ path (str): local path of the current evaluation image
+ names (dict(int, str)): hash map that maps class ids to labels
+ """
+ class_set = wandb.Classes([{'id': id, 'name': name} for id, name in names.items()])
+ box_data = []
+ avg_conf_per_class = [0] * len(self.data_dict['names'])
+ pred_class_count = {}
+ for *xyxy, conf, cls in predn.tolist():
+ if conf >= 0.25:
+ cls = int(cls)
+ box_data.append(
+ {"position": {"minX": xyxy[0], "minY": xyxy[1], "maxX": xyxy[2], "maxY": xyxy[3]},
+ "class_id": cls,
+ "box_caption": f"{names[cls]} {conf:.3f}",
+ "scores": {"class_score": conf},
+ "domain": "pixel"})
+ avg_conf_per_class[cls] += conf
+
+ if cls in pred_class_count:
+ pred_class_count[cls] += 1
+ else:
+ pred_class_count[cls] = 1
+
+ for pred_class in pred_class_count.keys():
+ avg_conf_per_class[pred_class] = avg_conf_per_class[pred_class] / pred_class_count[pred_class]
+
+ boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space
+ id = self.val_table_path_map[Path(path).name]
+ self.result_table.add_data(self.current_epoch,
+ id,
+ self.val_table.data[id][1],
+ wandb.Image(self.val_table.data[id][1], boxes=boxes, classes=class_set),
+ *avg_conf_per_class
+ )
+
+ def val_one_image(self, pred, predn, path, names, im):
+ """
+ Log validation data for one image. updates the result Table if validation dataset is uploaded and log bbox media panel
+
+ arguments:
+ pred (list): list of scaled predictions in the format - [xmin, ymin, xmax, ymax, confidence, class]
+ predn (list): list of predictions in the native space - [xmin, ymin, xmax, ymax, confidence, class]
+ path (str): local path of the current evaluation image
+ """
+ if self.val_table and self.result_table: # Log Table if Val dataset is uploaded as artifact
+ self.log_training_progress(predn, path, names)
+
+ if len(self.bbox_media_panel_images) < self.max_imgs_to_log and self.current_epoch > 0:
+ if self.current_epoch % self.bbox_interval == 0:
+ box_data = [{"position": {"minX": xyxy[0], "minY": xyxy[1], "maxX": xyxy[2], "maxY": xyxy[3]},
+ "class_id": int(cls),
+ "box_caption": f"{names[cls]} {conf:.3f}",
+ "scores": {"class_score": conf},
+ "domain": "pixel"} for *xyxy, conf, cls in pred.tolist()]
+ boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space
+ self.bbox_media_panel_images.append(wandb.Image(im, boxes=boxes, caption=path.name))
+
+ def log(self, log_dict):
+ """
+ save the metrics to the logging dictionary
+
+ arguments:
+ log_dict (Dict) -- metrics/media to be logged in current step
+ """
+ if self.wandb_run:
+ for key, value in log_dict.items():
+ self.log_dict[key] = value
+
+ def end_epoch(self, best_result=False):
+ """
+ commit the log_dict, model artifacts and Tables to W&B and flush the log_dict.
+
+ arguments:
+ best_result (boolean): Boolean representing if the result of this evaluation is best or not
+ """
+ if self.wandb_run:
+ with all_logging_disabled():
+ if self.bbox_media_panel_images:
+ self.log_dict["BoundingBoxDebugger"] = self.bbox_media_panel_images
+ try:
+ wandb.log(self.log_dict)
+ except BaseException as e:
+ LOGGER.info(
+ f"An error occurred in wandb logger. The training will proceed without interruption. More info\n{e}")
+ self.wandb_run.finish()
+ self.wandb_run = None
+
+ self.log_dict = {}
+ self.bbox_media_panel_images = []
+ if self.result_artifact:
+ self.result_artifact.add(self.result_table, 'result')
+ wandb.log_artifact(self.result_artifact, aliases=['latest', 'last', 'epoch ' + str(self.current_epoch),
+ ('best' if best_result else '')])
+
+ wandb.log({"evaluation": self.result_table})
+ columns = ["epoch", "id", "ground truth", "prediction"]
+ columns.extend(self.data_dict['names'])
+ self.result_table = wandb.Table(columns)
+ self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
+
+ def finish_run(self):
+ """
+ Log metrics if any and finish the current W&B run
+ """
+ if self.wandb_run:
+ if self.log_dict:
+ with all_logging_disabled():
+ wandb.log(self.log_dict)
+ wandb.run.finish()
+
+
+@contextmanager
+def all_logging_disabled(highest_level=logging.CRITICAL):
+ """ source - https://gist.github.com/simon-weber/7853144
+ A context manager that will prevent any logging messages triggered during the body from being processed.
+ :param highest_level: the maximum logging level in use.
+ This would only need to be changed if a custom level greater than CRITICAL is defined.
+ """
+ previous_level = logging.root.manager.disable
+ logging.disable(highest_level)
+ try:
+ yield
+ finally:
+ logging.disable(previous_level)
diff --git a/utils/loss.py b/utils/loss.py
new file mode 100644
index 0000000..45cee3e
--- /dev/null
+++ b/utils/loss.py
@@ -0,0 +1,662 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Loss functions
+"""
+
+import torch
+import torch.nn as nn
+
+from utils.metrics import bbox_iou
+from utils.torch_utils import is_parallel
+
+import torch.nn.functional as F
+
+from utils.general import box_iou, xywh2xyxy
+
+
+def smooth_BCE(eps=0.1): # https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441
+ # return positive, negative label smoothing BCE targets
+ return 1.0 - 0.5 * eps, 0.5 * eps
+
+
+class BCEBlurWithLogitsLoss(nn.Module):
+ # BCEwithLogitLoss() with reduced missing label effects.
+ def __init__(self, alpha=0.05):
+ super().__init__()
+ self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none') # must be nn.BCEWithLogitsLoss()
+ self.alpha = alpha
+
+ def forward(self, pred, true):
+ loss = self.loss_fcn(pred, true)
+ pred = torch.sigmoid(pred) # prob from logits
+ dx = pred - true # reduce only missing label effects
+ # dx = (pred - true).abs() # reduce missing label and false label effects
+ alpha_factor = 1 - torch.exp((dx - 1) / (self.alpha + 1e-4))
+ loss *= alpha_factor
+ return loss.mean()
+
+
+class FocalLoss(nn.Module):
+ # Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
+ def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
+ super().__init__()
+ self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss()
+ self.gamma = gamma
+ self.alpha = alpha
+ self.reduction = loss_fcn.reduction
+ self.loss_fcn.reduction = 'none' # required to apply FL to each element
+
+ def forward(self, pred, true):
+ loss = self.loss_fcn(pred, true)
+ # p_t = torch.exp(-loss)
+ # loss *= self.alpha * (1.000001 - p_t) ** self.gamma # non-zero power for gradient stability
+
+ # TF implementation https://github.com/tensorflow/addons/blob/v0.7.1/tensorflow_addons/losses/focal_loss.py
+ pred_prob = torch.sigmoid(pred) # prob from logits
+ p_t = true * pred_prob + (1 - true) * (1 - pred_prob)
+ alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
+ modulating_factor = (1.0 - p_t) ** self.gamma
+ loss *= alpha_factor * modulating_factor
+
+ if self.reduction == 'mean':
+ return loss.mean()
+ elif self.reduction == 'sum':
+ return loss.sum()
+ else: # 'none'
+ return loss
+
+
+class QFocalLoss(nn.Module):
+ # Wraps Quality focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
+ def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
+ super().__init__()
+ self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss()
+ self.gamma = gamma
+ self.alpha = alpha
+ self.reduction = loss_fcn.reduction
+ self.loss_fcn.reduction = 'none' # required to apply FL to each element
+
+ def forward(self, pred, true):
+ loss = self.loss_fcn(pred, true)
+
+ pred_prob = torch.sigmoid(pred) # prob from logits
+ alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
+ modulating_factor = torch.abs(true - pred_prob) ** self.gamma
+ loss *= alpha_factor * modulating_factor
+
+ if self.reduction == 'mean':
+ return loss.mean()
+ elif self.reduction == 'sum':
+ return loss.sum()
+ else: # 'none'
+ return loss
+
+
+class ComputeLoss:
+ # Compute losses
+ def __init__(self, model, autobalance=False):
+ self.sort_obj_iou = False
+ device = next(model.parameters()).device # get model device
+ h = model.hyp # hyperparameters
+
+ # Define criteria
+ BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))
+ BCEtheta = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['theta_pw']], device=device))
+ BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device))
+
+ # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
+ self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0)) # positive, negative BCE targets
+
+ # Focal loss
+ g = h['fl_gamma'] # focal loss gamma
+ if g > 0:
+ BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
+ BCEtheta = FocalLoss(BCEtheta, g)
+
+ det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module
+ self.stride = det.stride # tensor([8., 16., 32., ...])
+ self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, 0.02]) # P3-P7
+ # self.ssi = list(det.stride).index(16) if autobalance else 0 # stride 16 index
+ self.ssi = list(self.stride).index(16) if autobalance else 0 # stride 16 index
+ self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, 1.0, h, autobalance
+ self.BCEtheta = BCEtheta
+ for k in 'na', 'nc', 'nl', 'anchors':
+ setattr(self, k, getattr(det, k))
+
+ def __call__(self, p, targets): # predictions, targets, model
+ """
+ Args:
+ p (list[P3_out,...]): torch.Size(b, self.na, h_i, w_i, self.no), self.na means the number of anchors scales
+ targets (tensor): (n_gt_all_batch, [img_index clsid cx cy l s theta gaussian_θ_labels])
+
+ Return:
+ total_loss * bs (tensor): [1]
+ torch.cat((lbox, lobj, lcls, ltheta)).detach(): [4]
+ """
+ device = targets.device
+ lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device)
+ ltheta = torch.zeros(1, device=device)
+ # tcls, tbox, indices, anchors = self.build_targets(p, targets) # targets
+ tcls, tbox, indices, anchors, tgaussian_theta = self.build_targets(p, targets) # targets
+
+ # Losses
+ for i, pi in enumerate(p): # layer index, layer predictions
+ b, a, gj, gi = indices[i] # image, anchor, gridy, gridx
+ tobj = torch.zeros_like(pi[..., 0], device=device) # target obj
+
+ n = b.shape[0] # number of targets
+ if n:
+ ps = pi[b, a, gj, gi] # prediction subset corresponding to targets, (n_targets, self.no)
+
+ # Regression
+ pxy = ps[:, :2].sigmoid() * 2 - 0.5
+ pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i] # featuremap pixel
+ pbox = torch.cat((pxy, pwh), 1) # predicted box
+ iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True) # iou(prediction, target)
+ lbox += (1.0 - iou).mean() # iou loss
+
+ # Objectness
+ score_iou = iou.detach().clamp(0).type(tobj.dtype)
+ if self.sort_obj_iou:
+ sort_id = torch.argsort(score_iou)
+ b, a, gj, gi, score_iou = b[sort_id], a[sort_id], gj[sort_id], gi[sort_id], score_iou[sort_id]
+ tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * score_iou # iou ratio
+
+ # Classification
+ class_index = 5 + self.nc
+ if self.nc > 1: # cls loss (only if multiple classes)
+ # t = torch.full_like(ps[:, 5:], self.cn, device=device) # targets
+ t = torch.full_like(ps[:, 5:class_index], self.cn, device=device) # targets
+ t[range(n), tcls[i]] = self.cp
+ # lcls += self.BCEcls(ps[:, 5:], t) # BCE
+ lcls += self.BCEcls(ps[:, 5:class_index], t) # BCE
+
+ # theta Classification by Circular Smooth Label
+ t_theta = tgaussian_theta[i].type(ps.dtype) # target theta_gaussian_labels
+ ltheta += self.BCEtheta(ps[:, class_index:], t_theta)
+
+ # Append targets to text file
+ # with open('targets.txt', 'a') as file:
+ # [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]
+
+ obji = self.BCEobj(pi[..., 4], tobj)
+ lobj += obji * self.balance[i] # obj loss
+
+ if self.autobalance:
+ self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item()
+
+ if self.autobalance:
+ self.balance = [x / self.balance[self.ssi] for x in self.balance]
+ lbox *= self.hyp['box']
+ lobj *= self.hyp['obj']
+ lcls *= self.hyp['cls']
+ ltheta *= self.hyp['theta']
+ bs = tobj.shape[0] # batch size
+
+ # return (lbox + lobj + lcls) * bs, torch.cat((lbox, lobj, lcls)).detach()
+ return (lbox + lobj + lcls + ltheta) * bs, torch.cat((lbox, lobj, lcls, ltheta)).detach()
+
+ def build_targets(self, p, targets):
+ """
+ Args:
+ p (list[P3_out,...]): torch.Size(b, self.na, h_i, w_i, self.no), self.na means the number of anchors scales
+ targets (tensor): (n_gt_all_batch, [img_index clsid cx cy l s theta gaussian_θ_labels]) pixel
+
+ Return:non-normalized data
+ tcls (list[P3_out,...]): len=self.na, tensor.size(n_filter2)
+ tbox (list[P3_out,...]): len=self.na, tensor.size(n_filter2, 4) featuremap pixel
+ indices (list[P3_out,...]): len=self.na, tensor.size(4, n_filter2) [b, a, gj, gi]
+ anch (list[P3_out,...]): len=self.na, tensor.size(n_filter2, 2)
+ tgaussian_theta (list[P3_out,...]): len=self.na, tensor.size(n_filter2, hyp['cls_theta'])
+ # ttheta (list[P3_out,...]): len=self.na, tensor.size(n_filter2)
+ """
+ # Build targets for compute_loss()
+ na, nt = self.na, targets.shape[0] # number of anchors, targets
+ tcls, tbox, indices, anch = [], [], [], []
+ # ttheta, tgaussian_theta = [], []
+ tgaussian_theta = []
+ # gain = torch.ones(7, device=targets.device) # normalized to gridspace gain
+ feature_wh = torch.ones(2, device=targets.device) # feature_wh
+ ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt)
+ # targets (tensor): (n_gt_all_batch, c) -> (na, n_gt_all_batch, c) -> (na, n_gt_all_batch, c+1)
+ # targets (tensor): (na, n_gt_all_batch, [img_index, clsid, cx, cy, l, s, theta, gaussian_θ_labels, anchor_index]])
+ targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices
+
+ g = 0.5 # bias
+ off = torch.tensor([[0, 0], # tensor: (5, 2)
+ [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
+ # [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm
+ ], device=targets.device).float() * g # offsets
+
+ for i in range(self.nl):
+ anchors = self.anchors[i]
+ # gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain=[1, 1, w, h, w, h, 1, 1]
+ feature_wh[0:2] = torch.tensor(p[i].shape)[[3, 2]] # xyxy gain=[w_f, h_f]
+
+ # Match targets to anchors
+ # t = targets * gain # xywh featuremap pixel
+ t = targets.clone() # (na, n_gt_all_batch, c+1)
+ t[:, :, 2:6] /= self.stride[i] # xyls featuremap pixel
+ if nt:
+ # Matches
+ r = t[:, :, 4:6] / anchors[:, None] # edge_ls ratio, torch.size(na, n_gt_all_batch, 2)
+ j = torch.max(r, 1 / r).max(2)[0] < self.hyp['anchor_t'] # compare, torch.size(na, n_gt_all_batch)
+ # j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
+ t = t[j] # filter; Tensor.size(n_filter1, c+1)
+
+ # Offsets
+ gxy = t[:, 2:4] # grid xy; (n_filter1, 2)
+ # gxi = gain[[2, 3]] - gxy # inverse
+ gxi = feature_wh[[0, 1]] - gxy # inverse
+ j, k = ((gxy % 1 < g) & (gxy > 1)).T
+ l, m = ((gxi % 1 < g) & (gxi > 1)).T
+ j = torch.stack((torch.ones_like(j), j, k, l, m)) # (5, n_filter1)
+ t = t.repeat((5, 1, 1))[j] # (n_filter1, c+1) -> (5, n_filter1, c+1) -> (n_filter2, c+1)
+ offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j] # (5, n_filter1, 2) -> (n_filter2, 2)
+ else:
+ t = targets[0] # (n_gt_all_batch, c+1)
+ offsets = 0
+
+ # Define, t (tensor): (n_filter2, [img_index, clsid, cx, cy, l, s, theta, gaussian_θ_labels, anchor_index])
+ b, c = t[:, :2].long().T # image, class; (n_filter2)
+ gxy = t[:, 2:4] # grid xy
+ gwh = t[:, 4:6] # grid wh
+ # theta = t[:, 6]
+ gaussian_theta_labels = t[:, 7:-1]
+ gij = (gxy - offsets).long()
+ gi, gj = gij.T # grid xy indices
+
+ # Append
+ a = t[:, -1].long() # anchor indices 取整
+ # indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices
+ indices.append((b, a, gj.clamp_(0, feature_wh[1] - 1), gi.clamp_(0, feature_wh[0] - 1))) # image, anchor, grid indices
+ tbox.append(torch.cat((gxy - gij, gwh), 1)) # box
+ anch.append(anchors[a]) # anchors
+ tcls.append(c) # class
+ # ttheta.append(theta) # theta, θ∈[-pi/2, pi/2)
+ tgaussian_theta.append(gaussian_theta_labels)
+
+ # return tcls, tbox, indices, anch
+ return tcls, tbox, indices, anch, tgaussian_theta #, ttheta
+
+
+class ComputeLossOTA:
+ # Compute losses
+ def __init__(self, model, autobalance=False):
+ self.sort_obj_iou = False
+ device = next(model.parameters()).device # get model device
+ h = model.hyp # hyperparameters
+
+ # Define criteria
+ BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))
+ BCEtheta = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['theta_pw']], device=device))
+ BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device))
+
+ # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
+ self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0)) # positive, negative BCE targets
+
+ # Focal loss
+ g = h['fl_gamma'] # focal loss gamma
+ if g > 0:
+ BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
+ BCEtheta = FocalLoss(BCEtheta, g)
+
+ det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module
+ self.stride = det.stride # tensor([8., 16., 32., ...])
+ self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, 0.02]) # P3-P7
+ # self.ssi = list(det.stride).index(16) if autobalance else 0 # stride 16 index
+ self.ssi = list(self.stride).index(16) if autobalance else 0 # stride 16 index
+ self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, 1.0, h, autobalance
+ self.BCEtheta = BCEtheta
+ for k in 'na', 'nc', 'nl', 'anchors':
+ setattr(self, k, getattr(det, k))
+
+ def __call__(self, p, targets, imgs): # predictions, targets, model
+ device = targets.device
+ lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device)
+ ltheta = torch.zeros(1, device=device)
+ bs, as_, gjs, gis, targets, anchors, tgaussian_theta, indices = self.build_targets(p, targets, imgs)
+ pre_gen_gains = [torch.tensor(pp.shape, device=device)[[3, 2, 3, 2]] for pp in p]
+
+ # Losses
+ for i, pi in enumerate(p): # layer index, layer predictions
+ b, a, gj, gi = bs[i], as_[i], gjs[i], gis[i] # image, anchor, gridy, gridx
+ b1, a1, gj1, gi1 = indices[i]
+ tobj = torch.zeros_like(pi[..., 0], device=device) # target obj
+
+ n = b.shape[0] # number of targets
+ if n:
+ ps = pi[b, a, gj, gi] # prediction subset corresponding to targets
+ ps1 = pi[b1, a1, gj1, gi1]
+
+ # Regression
+ grid = torch.stack([gi, gj], dim=1)
+ pxy = ps[:, :2].sigmoid() * 2. - 0.5
+ # pxy = ps[:, :2].sigmoid() * 3. - 1.
+ pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i]
+ pbox = torch.cat((pxy, pwh), 1) # predicted box
+ selected_tbox = targets[i][:, 2:6] * pre_gen_gains[i]
+ selected_tbox[:, :2] -= grid
+ iou = bbox_iou(pbox.T, selected_tbox, x1y1x2y2=False, CIoU=True) # iou(prediction, target)
+ lbox += (1.0 - iou).mean() # iou loss
+
+ # Objectness
+ score_iou = iou.detach().clamp(0).type(tobj.dtype)
+ if self.sort_obj_iou:
+ sort_id = torch.argsort(score_iou)
+ b, a, gj, gi, score_iou = b[sort_id], a[sort_id], gj[sort_id], gi[sort_id], score_iou[sort_id]
+ tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype) # iou ratio
+
+ # Classification
+ selected_tcls = targets[i][:, 1].long()
+ class_index = 5 + self.nc
+ if self.nc > 1: # cls loss (only if multiple classes)
+ '''t = torch.full_like(ps[:, 5:], self.cn, device=device) # targets
+ t[range(n), selected_tcls] = self.cp
+ lcls += self.BCEcls(ps[:, 5:], t) # BCE'''
+
+ t = torch.full_like(ps[:, 5:class_index], self.cn, device=device) # targets
+ t[range(n), selected_tcls] = self.cp
+ lcls += self.BCEcls(ps[:, 5:class_index], t) # BCE
+
+ # Append targets to text file
+ # with open('targets.txt', 'a') as file:
+ # [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]
+
+ t_theta = tgaussian_theta[i].type(ps1.dtype) # target theta_gaussian_labels
+ ltheta += self.BCEtheta(ps1[:, class_index:], t_theta)
+
+ obji = self.BCEobj(pi[..., 4], tobj)
+ lobj += obji * self.balance[i] # obj loss
+ if self.autobalance:
+ self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item()
+
+ if self.autobalance:
+ self.balance = [x / self.balance[self.ssi] for x in self.balance]
+ lbox *= self.hyp['box']
+ lobj *= self.hyp['obj']
+ lcls *= self.hyp['cls']
+ ltheta *= self.hyp['theta']
+ bs = tobj.shape[0] # batch size
+
+ loss = lbox + lobj + lcls + ltheta
+ #return (lbox + lobj + lcls + ltheta) * bs, torch.cat((lbox, lobj, lcls, ltheta)).detach()
+ return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach()
+
+ def build_targets(self, p, targets, imgs):
+
+ # indices, anch = self.find_positive(p, targets)
+ indices, anch, tgaussian_theta = self.find_3_positive(p, targets)
+ # indices, anch = self.find_4_positive(p, targets)
+ # indices, anch = self.find_5_positive(p, targets)
+ # indices, anch = self.find_9_positive(p, targets)
+
+ matching_bs = [[] for pp in p]
+ matching_as = [[] for pp in p]
+ matching_gjs = [[] for pp in p]
+ matching_gis = [[] for pp in p]
+ matching_targets = [[] for pp in p]
+ matching_anchs = [[] for pp in p]
+
+ nl = len(p)
+
+ for batch_idx in range(p[0].shape[0]):
+
+ b_idx = targets[:, 0] == batch_idx
+ this_target = targets[b_idx]
+ if this_target.shape[0] == 0:
+ continue
+
+ txywh = this_target[:, 2:6] * imgs[batch_idx].shape[1]
+ txyxy = xywh2xyxy(txywh)
+
+ pxyxys = []
+ p_cls = []
+ p_obj = []
+ from_which_layer = []
+ all_b = []
+ all_a = []
+ all_gj = []
+ all_gi = []
+ all_anch = []
+
+ for i, pi in enumerate(p):
+ b, a, gj, gi = indices[i]
+ idx = (b == batch_idx)
+ b, a, gj, gi = b[idx], a[idx], gj[idx], gi[idx]
+ all_b.append(b)
+ all_a.append(a)
+ all_gj.append(gj)
+ all_gi.append(gi)
+ all_anch.append(anch[i][idx])
+ from_which_layer.append(torch.ones(size=(len(b),)) * i)
+
+ fg_pred = pi[b, a, gj, gi]
+ p_obj.append(fg_pred[:, 4:5])
+ p_cls.append(fg_pred[:, 5:])
+
+ grid = torch.stack([gi, gj], dim=1)
+ pxy = (fg_pred[:, :2].sigmoid() * 2. - 0.5 + grid) * self.stride[i] # / 8.
+ # pxy = (fg_pred[:, :2].sigmoid() * 3. - 1. + grid) * self.stride[i]
+ pwh = (fg_pred[:, 2:4].sigmoid() * 2) ** 2 * anch[i][idx] * self.stride[i] # / 8.
+ pxywh = torch.cat([pxy, pwh], dim=-1)
+ pxyxy = xywh2xyxy(pxywh)
+ pxyxys.append(pxyxy)
+
+ pxyxys = torch.cat(pxyxys, dim=0)
+ if pxyxys.shape[0] == 0:
+ continue
+ p_obj = torch.cat(p_obj, dim=0)
+ p_cls = torch.cat(p_cls, dim=0)
+ from_which_layer = torch.cat(from_which_layer, dim=0)
+ all_b = torch.cat(all_b, dim=0)
+ all_a = torch.cat(all_a, dim=0)
+ all_gj = torch.cat(all_gj, dim=0)
+ all_gi = torch.cat(all_gi, dim=0)
+ all_anch = torch.cat(all_anch, dim=0)
+
+ pair_wise_iou = box_iou(txyxy, pxyxys)
+
+ pair_wise_iou_loss = -torch.log(pair_wise_iou + 1e-8)
+
+ top_k, _ = torch.topk(pair_wise_iou, min(10, pair_wise_iou.shape[1]), dim=1)
+ dynamic_ks = torch.clamp(top_k.sum(1).int(), min=1)
+
+ gt_cls_per_image = (
+ F.one_hot(this_target[:, 1].to(torch.int64), self.nc)
+ .float()
+ .unsqueeze(1)
+ .repeat(1, pxyxys.shape[0], 181)
+ )
+
+ num_gt = this_target.shape[0]
+
+ cls_preds_ = (
+ p_cls.float().unsqueeze(0).repeat(num_gt, 1, 1).sigmoid_()
+ * p_obj.unsqueeze(0).repeat(num_gt, 1, 1).sigmoid_()
+ )
+
+ y = cls_preds_.sqrt_()
+ pair_wise_cls_loss = F.binary_cross_entropy_with_logits(
+ torch.log(y / (1 - y)), gt_cls_per_image, reduction="none"
+ ).sum(-1)
+ del cls_preds_
+
+ cost = (
+ pair_wise_cls_loss
+ + 3.0 * pair_wise_iou_loss
+ )
+
+ matching_matrix = torch.zeros_like(cost)
+
+ for gt_idx in range(num_gt):
+ _, pos_idx = torch.topk(
+ cost[gt_idx], k=dynamic_ks[gt_idx].item(), largest=False
+ )
+ matching_matrix[gt_idx][pos_idx] = 1.0
+
+ del top_k, dynamic_ks
+ anchor_matching_gt = matching_matrix.sum(0)
+ if (anchor_matching_gt > 1).sum() > 0:
+ _, cost_argmin = torch.min(cost[:, anchor_matching_gt > 1], dim=0)
+ matching_matrix[:, anchor_matching_gt > 1] *= 0.0
+ matching_matrix[cost_argmin, anchor_matching_gt > 1] = 1.0
+ fg_mask_inboxes = matching_matrix.sum(0) > 0.0
+ matched_gt_inds = matching_matrix[:, fg_mask_inboxes].argmax(0)
+
+ from_which_layer = from_which_layer[fg_mask_inboxes]
+ all_b = all_b[fg_mask_inboxes]
+ all_a = all_a[fg_mask_inboxes]
+ all_gj = all_gj[fg_mask_inboxes]
+ all_gi = all_gi[fg_mask_inboxes]
+ all_anch = all_anch[fg_mask_inboxes]
+
+ this_target = this_target[matched_gt_inds]
+
+ for i in range(nl):
+ layer_idx = from_which_layer == i
+ matching_bs[i].append(all_b[layer_idx])
+ matching_as[i].append(all_a[layer_idx])
+ matching_gjs[i].append(all_gj[layer_idx])
+ matching_gis[i].append(all_gi[layer_idx])
+ matching_targets[i].append(this_target[layer_idx])
+ matching_anchs[i].append(all_anch[layer_idx])
+
+ for i in range(nl):
+ if matching_targets[i] != []:
+ matching_bs[i] = torch.cat(matching_bs[i], dim=0)
+ matching_as[i] = torch.cat(matching_as[i], dim=0)
+ matching_gjs[i] = torch.cat(matching_gjs[i], dim=0)
+ matching_gis[i] = torch.cat(matching_gis[i], dim=0)
+ matching_targets[i] = torch.cat(matching_targets[i], dim=0)
+ matching_anchs[i] = torch.cat(matching_anchs[i], dim=0)
+ else:
+ matching_bs[i] = torch.tensor([], device='cuda:0', dtype=torch.int64)
+ matching_as[i] = torch.tensor([], device='cuda:0', dtype=torch.int64)
+ matching_gjs[i] = torch.tensor([], device='cuda:0', dtype=torch.int64)
+ matching_gis[i] = torch.tensor([], device='cuda:0', dtype=torch.int64)
+ matching_targets[i] = torch.tensor([], device='cuda:0', dtype=torch.int64)
+ matching_anchs[i] = torch.tensor([], device='cuda:0', dtype=torch.int64)
+
+ return matching_bs, matching_as, matching_gjs, matching_gis, matching_targets, matching_anchs, tgaussian_theta, indices
+
+ def find_3_positive(self, p, targets):
+ # Build targets for compute_loss(), input targets(image,class,x,y,w,h)
+ '''na, nt = self.na, targets.shape[0] # number of anchors, targets
+ indices, anch = [], []
+ gain = torch.ones(7, device=targets.device).long() # normalized to gridspace gain
+ ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt)
+ targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices
+
+ g = 0.5 # bias
+ off = torch.tensor([[0, 0],
+ [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
+ # [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm
+ ], device=targets.device).float() * g # offsets
+
+ for i in range(self.nl):
+ anchors = self.anchors[i]
+ gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain
+
+ # Match targets to anchors
+ t = targets * gain
+ if nt:
+ # Matches
+ r = t[:, :, 4:6] / anchors[:, None] # wh ratio
+ j = torch.max(r, 1. / r).max(2)[0] < self.hyp['anchor_t'] # compare
+ # j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
+ t = t[j] # filter
+
+ # Offsets
+ gxy = t[:, 2:4] # grid xy
+ gxi = gain[[2, 3]] - gxy # inverse
+ j, k = ((gxy % 1. < g) & (gxy > 1.)).T
+ l, m = ((gxi % 1. < g) & (gxi > 1.)).T
+ j = torch.stack((torch.ones_like(j), j, k, l, m))
+ t = t.repeat((5, 1, 1))[j]
+ offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
+ else:
+ t = targets[0]
+ offsets = 0
+
+ # Define
+ b, c = t[:, :2].long().T # image, class
+ gxy = t[:, 2:4] # grid xy
+ gwh = t[:, 4:6] # grid wh
+ gij = (gxy - offsets).long()
+ gi, gj = gij.T # grid xy indices
+
+ # Append
+ a = t[:, 6].long() # anchor indices
+ indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices
+ anch.append(anchors[a]) # anchors
+
+ return indices, anch'''
+ na, nt = self.na, targets.shape[0] # number of anchors, targets
+ tcls, tbox, indices, anch = [], [], [], []
+ # ttheta, tgaussian_theta = [], []
+ tgaussian_theta = []
+ # gain = torch.ones(7, device=targets.device) # normalized to gridspace gain
+ feature_wh = torch.ones(2, device=targets.device) # feature_wh
+ ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt)
+ # targets (tensor): (n_gt_all_batch, c) -> (na, n_gt_all_batch, c) -> (na, n_gt_all_batch, c+1)
+ # targets (tensor): (na, n_gt_all_batch, [img_index, clsid, cx, cy, l, s, theta, gaussian_θ_labels, anchor_index]])
+ targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices
+
+ g = 0.5 # bias
+ off = torch.tensor([[0, 0], # tensor: (5, 2)
+ [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
+ # [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm
+ ], device=targets.device).float() * g # offsets
+
+ for i in range(self.nl):
+ anchors = self.anchors[i]
+ # gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain=[1, 1, w, h, w, h, 1, 1]
+ feature_wh[0:2] = torch.tensor(p[i].shape)[[3, 2]] # xyxy gain=[w_f, h_f]
+
+ # Match targets to anchors
+ # t = targets * gain # xywh featuremap pixel
+ t = targets.clone() # (na, n_gt_all_batch, c+1)
+ t[:, :, 2:6] /= self.stride[i] # xyls featuremap pixel
+ if nt:
+ # Matches
+ r = t[:, :, 4:6] / anchors[:, None] # edge_ls ratio, torch.size(na, n_gt_all_batch, 2)
+ j = torch.max(r, 1 / r).max(2)[0] < self.hyp['anchor_t'] # compare, torch.size(na, n_gt_all_batch)
+ # j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
+ t = t[j] # filter; Tensor.size(n_filter1, c+1)
+
+ # Offsets
+ gxy = t[:, 2:4] # grid xy; (n_filter1, 2)
+ # gxi = gain[[2, 3]] - gxy # inverse
+ gxi = feature_wh[[0, 1]] - gxy # inverse
+ j, k = ((gxy % 1 < g) & (gxy > 1)).T
+ l, m = ((gxi % 1 < g) & (gxi > 1)).T
+ j = torch.stack((torch.ones_like(j), j, k, l, m)) # (5, n_filter1)
+ t = t.repeat((5, 1, 1))[j] # (n_filter1, c+1) -> (5, n_filter1, c+1) -> (n_filter2, c+1)
+ offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j] # (5, n_filter1, 2) -> (n_filter2, 2)
+ else:
+ t = targets[0] # (n_gt_all_batch, c+1)
+ offsets = 0
+
+ # Define, t (tensor): (n_filter2, [img_index, clsid, cx, cy, l, s, theta, gaussian_θ_labels, anchor_index])
+ b, c = t[:, :2].long().T # image, class; (n_filter2)
+ gxy = t[:, 2:4] # grid xy
+ gwh = t[:, 4:6] # grid wh
+ # theta = t[:, 6]
+ gaussian_theta_labels = t[:, 7:-1]
+ gij = (gxy - offsets).long()
+ gi, gj = gij.T # grid xy indices
+
+ # Append
+ a = t[:, -1].long() # anchor indices 取整
+ # indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices
+ indices.append(
+ (b, a, gj.clamp_(0, feature_wh[1] - 1), gi.clamp_(0, feature_wh[0] - 1))) # image, anchor, grid indices
+ tbox.append(torch.cat((gxy - gij, gwh), 1)) # box
+ anch.append(anchors[a]) # anchors
+ tcls.append(c) # class
+ # ttheta.append(theta) # theta, θ∈[-pi/2, pi/2)
+ tgaussian_theta.append(gaussian_theta_labels)
+
+ # return tcls, tbox, indices, anch
+ return indices, anch, tgaussian_theta # , ttheta
\ No newline at end of file
diff --git a/utils/metrics.py b/utils/metrics.py
new file mode 100644
index 0000000..c80bfa2
--- /dev/null
+++ b/utils/metrics.py
@@ -0,0 +1,344 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Model validation metrics
+"""
+
+import math
+import warnings
+from pathlib import Path
+
+import matplotlib.pyplot as plt
+import numpy as np
+import torch
+
+
+def fitness(x):
+ # Model fitness as a weighted combination of metrics
+ w = [0.0, 0.0, 0.1, 0.9] # weights for [P, R, mAP@0.5, mAP@0.5:0.95]
+ return (x[:, :4] * w).sum(1)
+
+
+def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=(), eps=1e-16):
+ """ Compute the average precision, given the recall and precision curves.
+ Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
+ # Arguments
+ tp: True positives (nparray, nx1 or nx10).
+ conf: Objectness value from 0-1 (nparray).
+ pred_cls: Predicted object classes (nparray).
+ target_cls: True object classes (nparray).
+ plot: Plot precision-recall curve at mAP@0.5
+ save_dir: Plot save directory
+ # Returns
+ The average precision as computed in py-faster-rcnn.
+ """
+
+ # Sort by objectness
+ i = np.argsort(-conf)
+ tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]
+
+ # Find unique classes
+ unique_classes, nt = np.unique(target_cls, return_counts=True)
+ nc = unique_classes.shape[0] # number of classes, number of detections
+
+ # Create Precision-Recall curve and compute AP for each class
+ px, py = np.linspace(0, 1, 1000), [] # for plotting
+ ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000))
+ for ci, c in enumerate(unique_classes):
+ i = pred_cls == c
+ n_l = nt[ci] # number of labels
+ n_p = i.sum() # number of predictions
+
+ if n_p == 0 or n_l == 0:
+ continue
+ else:
+ # Accumulate FPs and TPs
+ fpc = (1 - tp[i]).cumsum(0)
+ tpc = tp[i].cumsum(0)
+
+ # Recall
+ recall = tpc / (n_l + eps) # recall curve
+ r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0) # negative x, xp because xp decreases
+
+ # Precision
+ precision = tpc / (tpc + fpc) # precision curve
+ p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1) # p at pr_score
+
+ # AP from recall-precision curve
+ for j in range(tp.shape[1]):
+ ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j])
+ if plot and j == 0:
+ py.append(np.interp(px, mrec, mpre)) # precision at mAP@0.5
+
+ # Compute F1 (harmonic mean of precision and recall)
+ f1 = 2 * p * r / (p + r + eps)
+ names = [v for k, v in names.items() if k in unique_classes] # list: only classes that have data
+ names = {i: v for i, v in enumerate(names)} # to dict
+ if plot:
+ plot_pr_curve(px, py, ap, Path(save_dir) / 'PR_curve.png', names)
+ plot_mc_curve(px, f1, Path(save_dir) / 'F1_curve.png', names, ylabel='F1')
+ plot_mc_curve(px, p, Path(save_dir) / 'P_curve.png', names, ylabel='Precision')
+ plot_mc_curve(px, r, Path(save_dir) / 'R_curve.png', names, ylabel='Recall')
+
+ i = f1.mean(0).argmax() # max F1 index
+ p, r, f1 = p[:, i], r[:, i], f1[:, i]
+ tp = (r * nt).round() # true positives
+ fp = (tp / (p + eps) - tp).round() # false positives
+ return tp, fp, p, r, f1, ap, unique_classes.astype('int32')
+
+
+def compute_ap(recall, precision):
+ """ Compute the average precision, given the recall and precision curves
+ # Arguments
+ recall: The recall curve (list)
+ precision: The precision curve (list)
+ # Returns
+ Average precision, precision curve, recall curve
+ """
+
+ # Append sentinel values to beginning and end
+ mrec = np.concatenate(([0.0], recall, [1.0]))
+ mpre = np.concatenate(([1.0], precision, [0.0]))
+
+ # Compute the precision envelope
+ mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))
+
+ # Integrate area under curve
+ method = 'interp' # methods: 'continuous', 'interp'
+ if method == 'interp':
+ x = np.linspace(0, 1, 101) # 101-point interp (COCO)
+ ap = np.trapz(np.interp(x, mrec, mpre), x) # integrate
+ else: # 'continuous'
+ i = np.where(mrec[1:] != mrec[:-1])[0] # points where x axis (recall) changes
+ ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) # area under curve
+
+ return ap, mpre, mrec
+
+
+class ConfusionMatrix:
+ # Updated version of https://github.com/kaanakan/object_detection_confusion_matrix
+ def __init__(self, nc, conf=0.25, iou_thres=0.45):
+ self.matrix = np.zeros((nc + 1, nc + 1))
+ self.nc = nc # number of classes
+ self.conf = conf
+ self.iou_thres = iou_thres
+
+ def process_batch(self, detections, labels):
+ """
+ Return intersection-over-union (Jaccard index) of boxes.
+ Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
+ Arguments:
+ detections (Array[N, 6]), x1, y1, x2, y2, conf, class
+ labels (Array[M, 5]), class, x1, y1, x2, y2
+ Returns:
+ None, updates confusion matrix accordingly
+ """
+ detections = detections[detections[:, 4] > self.conf]
+ gt_classes = labels[:, 0].int()
+ detection_classes = detections[:, 5].int()
+ iou = box_iou(labels[:, 1:], detections[:, :4])
+
+ x = torch.where(iou > self.iou_thres)
+ if x[0].shape[0]:
+ matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy()
+ if x[0].shape[0] > 1:
+ matches = matches[matches[:, 2].argsort()[::-1]]
+ matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
+ matches = matches[matches[:, 2].argsort()[::-1]]
+ matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
+ else:
+ matches = np.zeros((0, 3))
+
+ n = matches.shape[0] > 0
+ m0, m1, _ = matches.transpose().astype(np.int16)
+ for i, gc in enumerate(gt_classes):
+ j = m0 == i
+ if n and sum(j) == 1:
+ self.matrix[detection_classes[m1[j]], gc] += 1 # correct
+ else:
+ self.matrix[self.nc, gc] += 1 # background FP
+
+ if n:
+ for i, dc in enumerate(detection_classes):
+ if not any(m1 == i):
+ self.matrix[dc, self.nc] += 1 # background FN
+
+ def matrix(self):
+ return self.matrix
+
+ def tp_fp(self):
+ tp = self.matrix.diagonal() # true positives
+ fp = self.matrix.sum(1) - tp # false positives
+ # fn = self.matrix.sum(0) - tp # false negatives (missed detections)
+ return tp[:-1], fp[:-1] # remove background class
+
+ def plot(self, normalize=True, save_dir='', names=()):
+ try:
+ import seaborn as sn
+
+ array = self.matrix / ((self.matrix.sum(0).reshape(1, -1) + 1E-6) if normalize else 1) # normalize columns
+ array[array < 0.005] = np.nan # don't annotate (would appear as 0.00)
+
+ fig = plt.figure(figsize=(12, 9), tight_layout=True)
+ sn.set(font_scale=1.0 if self.nc < 50 else 0.8) # for label size
+ labels = (0 < len(names) < 99) and len(names) == self.nc # apply names to ticklabels
+ with warnings.catch_warnings():
+ warnings.simplefilter('ignore') # suppress empty matrix RuntimeWarning: All-NaN slice encountered
+ sn.heatmap(array, annot=self.nc < 30, annot_kws={"size": 8}, cmap='Blues', fmt='.2f', square=True,
+ xticklabels=names + ['background FP'] if labels else "auto",
+ yticklabels=names + ['background FN'] if labels else "auto").set_facecolor((1, 1, 1))
+ fig.axes[0].set_xlabel('True')
+ fig.axes[0].set_ylabel('Predicted')
+ fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250)
+ plt.close()
+ except Exception as e:
+ print(f'WARNING: ConfusionMatrix plot failure: {e}')
+
+ def print(self):
+ for i in range(self.nc + 1):
+ print(' '.join(map(str, self.matrix[i])))
+
+
+def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7):
+ # Returns the IoU of box1 to box2. box1 is 4, box2 is nx4
+ box2 = box2.T
+
+ # Get the coordinates of bounding boxes
+ if x1y1x2y2: # x1, y1, x2, y2 = box1
+ b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
+ b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
+ else: # transform from xywh to xyxy
+ b1_x1, b1_x2 = box1[0] - box1[2] / 2, box1[0] + box1[2] / 2
+ b1_y1, b1_y2 = box1[1] - box1[3] / 2, box1[1] + box1[3] / 2
+ b2_x1, b2_x2 = box2[0] - box2[2] / 2, box2[0] + box2[2] / 2
+ b2_y1, b2_y2 = box2[1] - box2[3] / 2, box2[1] + box2[3] / 2
+
+ # Intersection area
+ inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
+ (torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)
+
+ # Union Area
+ w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps
+ w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps
+ union = w1 * h1 + w2 * h2 - inter + eps
+
+ iou = inter / union
+ if CIoU or DIoU or GIoU:
+ cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1) # convex (smallest enclosing box) width
+ ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1) # convex height
+ if CIoU or DIoU: # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
+ c2 = cw ** 2 + ch ** 2 + eps # convex diagonal squared
+ rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 +
+ (b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4 # center distance squared
+ if CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
+ v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
+ with torch.no_grad():
+ alpha = v / (v - iou + (1 + eps))
+ return iou - (rho2 / c2 + v * alpha) # CIoU
+ else:
+ return iou - rho2 / c2 # DIoU
+ else: # GIoU https://arxiv.org/pdf/1902.09630.pdf
+ c_area = cw * ch + eps # convex area
+ return iou - (c_area - union) / c_area # GIoU
+ else:
+ return iou # IoU
+
+
+def box_iou(box1, box2):
+ # https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
+ """
+ Return intersection-over-union (Jaccard index) of boxes.
+ Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
+ Arguments:
+ box1 (Tensor[N, 4])
+ box2 (Tensor[M, 4])
+ Returns:
+ iou (Tensor[N, M]): the NxM matrix containing the pairwise
+ IoU values for every element in boxes1 and boxes2
+ """
+
+ def box_area(box):
+ # box = 4xn
+ return (box[2] - box[0]) * (box[3] - box[1])
+
+ area1 = box_area(box1.T)
+ area2 = box_area(box2.T)
+
+ # inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
+ inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
+ return inter / (area1[:, None] + area2 - inter) # iou = inter / (area1 + area2 - inter)
+
+
+def bbox_ioa(box1, box2, eps=1E-7):
+ """ Returns the intersection over box2 area given box1, box2. Boxes are x1y1x2y2
+ box1: np.array of shape(4)
+ box2: np.array of shape(nx4)
+ returns: np.array of shape(n)
+ """
+
+ box2 = box2.transpose()
+
+ # Get the coordinates of bounding boxes
+ b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
+ b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
+
+ # Intersection area
+ inter_area = (np.minimum(b1_x2, b2_x2) - np.maximum(b1_x1, b2_x1)).clip(0) * \
+ (np.minimum(b1_y2, b2_y2) - np.maximum(b1_y1, b2_y1)).clip(0)
+
+ # box2 area
+ box2_area = (b2_x2 - b2_x1) * (b2_y2 - b2_y1) + eps
+
+ # Intersection over box2 area
+ return inter_area / box2_area
+
+
+def wh_iou(wh1, wh2):
+ # Returns the nxm IoU matrix. wh1 is nx2, wh2 is mx2
+ wh1 = wh1[:, None] # [N,1,2]
+ wh2 = wh2[None] # [1,M,2]
+ inter = torch.min(wh1, wh2).prod(2) # [N,M]
+ return inter / (wh1.prod(2) + wh2.prod(2) - inter) # iou = inter / (area1 + area2 - inter)
+
+
+# Plots ----------------------------------------------------------------------------------------------------------------
+
+def plot_pr_curve(px, py, ap, save_dir='pr_curve.png', names=()):
+ # Precision-recall curve
+ fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)
+ py = np.stack(py, axis=1)
+
+ if 0 < len(names) < 21: # display per-class legend if < 21 classes
+ for i, y in enumerate(py.T):
+ ax.plot(px, y, linewidth=1, label=f'{names[i]} {ap[i, 0]:.3f}') # plot(recall, precision)
+ else:
+ ax.plot(px, py, linewidth=1, color='grey') # plot(recall, precision)
+
+ ax.plot(px, py.mean(1), linewidth=3, color='blue', label='all classes %.3f hbb mAP@0.5' % ap[:, 0].mean())
+ ax.set_xlabel('Recall')
+ ax.set_ylabel('Precision')
+ ax.set_xlim(0, 1)
+ ax.set_ylim(0, 1)
+ plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
+ fig.savefig(Path(save_dir), dpi=250)
+ plt.close()
+
+
+def plot_mc_curve(px, py, save_dir='mc_curve.png', names=(), xlabel='Confidence', ylabel='Metric'):
+ # Metric-confidence curve
+ fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)
+
+ if 0 < len(names) < 21: # display per-class legend if < 21 classes
+ for i, y in enumerate(py):
+ ax.plot(px, y, linewidth=1, label=f'{names[i]}') # plot(confidence, metric)
+ else:
+ ax.plot(px, py.T, linewidth=1, color='grey') # plot(confidence, metric)
+
+ y = py.mean(0)
+ ax.plot(px, y, linewidth=3, color='blue', label=f'all classes {y.max():.2f} at {px[y.argmax()]:.3f}')
+ ax.set_xlabel(xlabel)
+ ax.set_ylabel(ylabel)
+ ax.set_xlim(0, 1)
+ ax.set_ylim(0, 1)
+ plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
+ fig.savefig(Path(save_dir), dpi=250)
+ plt.close()
diff --git a/utils/nms_rotated/__init__.py b/utils/nms_rotated/__init__.py
new file mode 100644
index 0000000..9768d17
--- /dev/null
+++ b/utils/nms_rotated/__init__.py
@@ -0,0 +1,3 @@
+from .nms_rotated_wrapper import obb_nms, poly_nms
+
+__all__ = ['obb_nms', 'poly_nms']
diff --git a/utils/nms_rotated/__pycache__/__init__.cpython-37.pyc b/utils/nms_rotated/__pycache__/__init__.cpython-37.pyc
new file mode 100644
index 0000000..6ec2e44
Binary files /dev/null and b/utils/nms_rotated/__pycache__/__init__.cpython-37.pyc differ
diff --git a/utils/nms_rotated/__pycache__/nms_rotated_wrapper.cpython-37.pyc b/utils/nms_rotated/__pycache__/nms_rotated_wrapper.cpython-37.pyc
new file mode 100644
index 0000000..ef94b65
Binary files /dev/null and b/utils/nms_rotated/__pycache__/nms_rotated_wrapper.cpython-37.pyc differ
diff --git a/utils/nms_rotated/build/lib.win-amd64-3.7/nms_rotated_ext.pyd b/utils/nms_rotated/build/lib.win-amd64-3.7/nms_rotated_ext.pyd
new file mode 100644
index 0000000..ec6c224
Binary files /dev/null and b/utils/nms_rotated/build/lib.win-amd64-3.7/nms_rotated_ext.pyd differ
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/.ninja_deps b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/.ninja_deps
new file mode 100644
index 0000000..bb12204
Binary files /dev/null and b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/.ninja_deps differ
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/.ninja_log b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/.ninja_log
new file mode 100644
index 0000000..b0f0c7f
--- /dev/null
+++ b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/.ninja_log
@@ -0,0 +1,9 @@
+# ninja log v5
+2 3053 6788011957379629 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cpu.obj 230f582cc4ead385
+2 6945 6788011996267406 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.obj 91ac4870d9dfe29f
+3 3474 6789732946931915 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cpu.obj 1256c1d3ee98246c
+1 7591 6789732988080942 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.obj 62d6bb08f7474408
+3 3117 6789734007496664 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cpu.obj 230f582cc4ead385
+2 6897 6789734045285714 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.obj 91ac4870d9dfe29f
+3 8105 6789761451993078 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/poly_nms_cuda.obj de5e56148eafc714
+1 8722 6789761458150954 C:/Users/Administrator/Desktop/yolov5_obb/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cuda.obj 8782fd757258be9d
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/build.ninja b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/build.ninja
new file mode 100644
index 0000000..6700a49
--- /dev/null
+++ b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/build.ninja
@@ -0,0 +1,28 @@
+ninja_required_version = 1.3
+cxx = cl
+nvcc = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0\bin\nvcc
+
+cflags = /nologo /O2 /W3 /GL /DNDEBUG /MD /MD /wd4819 /wd4251 /wd4244 /wd4267 /wd4275 /wd4018 /wd4190 /EHsc -DWITH_CUDA -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include\torch\csrc\api\include -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include\TH -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0\include" -ID:\anaconda3\envs\pytorch\include -ID:\anaconda3\envs\pytorch\Include "-IC:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.29.30133\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\shared" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\um" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\winrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\cppwinrt"
+post_cflags = -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=nms_rotated_ext -D_GLIBCXX_USE_CXX11_ABI=0 /std:c++14
+cuda_cflags = --use-local-env -Xcompiler /MD -Xcompiler /wd4819 -Xcompiler /wd4251 -Xcompiler /wd4244 -Xcompiler /wd4267 -Xcompiler /wd4275 -Xcompiler /wd4018 -Xcompiler /wd4190 -Xcompiler /EHsc -Xcudafe --diag_suppress=base_class_has_different_dll_interface -Xcudafe --diag_suppress=field_without_dll_interface -Xcudafe --diag_suppress=dll_interface_conflict_none_assumed -Xcudafe --diag_suppress=dll_interface_conflict_dllexport_assumed -DWITH_CUDA -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include\torch\csrc\api\include -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include\TH -ID:\anaconda3\envs\pytorch\lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0\include" -ID:\anaconda3\envs\pytorch\include -ID:\anaconda3\envs\pytorch\Include "-IC:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.29.30133\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\shared" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\um" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\winrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\cppwinrt"
+cuda_post_cflags = -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=nms_rotated_ext -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_80,code=sm_80
+ldflags =
+
+rule compile
+ command = cl /showIncludes $cflags -c $in /Fo$out $post_cflags
+ deps = msvc
+
+rule cuda_compile
+ command = $nvcc $cuda_cflags -c $in -o $out $cuda_post_cflags
+
+
+
+build C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\build\temp.win-amd64-3.7\Release\src/nms_rotated_cpu.obj: compile C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\src\nms_rotated_cpu.cpp
+build C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\build\temp.win-amd64-3.7\Release\src/nms_rotated_cuda.obj: cuda_compile C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\src\nms_rotated_cuda.cu
+build C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\build\temp.win-amd64-3.7\Release\src/nms_rotated_ext.obj: compile C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\src\nms_rotated_ext.cpp
+build C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\build\temp.win-amd64-3.7\Release\src/poly_nms_cuda.obj: cuda_compile C$:\Users\Administrator\Desktop\yolov5_obb\utils\nms_rotated\src\poly_nms_cuda.cu
+
+
+
+
+
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cpu.obj b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cpu.obj
new file mode 100644
index 0000000..9e1c6d5
Binary files /dev/null and b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cpu.obj differ
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cuda.obj b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cuda.obj
new file mode 100644
index 0000000..3980701
Binary files /dev/null and b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_cuda.obj differ
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.exp b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.exp
new file mode 100644
index 0000000..b834ae3
Binary files /dev/null and b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.exp differ
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.lib b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.lib
new file mode 100644
index 0000000..ff8a16a
Binary files /dev/null and b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.lib differ
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.obj b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.obj
new file mode 100644
index 0000000..98c1c5f
Binary files /dev/null and b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/nms_rotated_ext.obj differ
diff --git a/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/poly_nms_cuda.obj b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/poly_nms_cuda.obj
new file mode 100644
index 0000000..d9ba8f4
Binary files /dev/null and b/utils/nms_rotated/build/temp.win-amd64-3.7/Release/src/poly_nms_cuda.obj differ
diff --git a/utils/nms_rotated/nms_rotated.egg-info/PKG-INFO b/utils/nms_rotated/nms_rotated.egg-info/PKG-INFO
new file mode 100644
index 0000000..83acf87
--- /dev/null
+++ b/utils/nms_rotated/nms_rotated.egg-info/PKG-INFO
@@ -0,0 +1,9 @@
+Metadata-Version: 2.1
+Name: nms-rotated
+Version: 0.0.0
+Summary: UNKNOWN
+License: UNKNOWN
+Platform: UNKNOWN
+
+UNKNOWN
+
diff --git a/utils/nms_rotated/nms_rotated.egg-info/SOURCES.txt b/utils/nms_rotated/nms_rotated.egg-info/SOURCES.txt
new file mode 100644
index 0000000..8e475c6
--- /dev/null
+++ b/utils/nms_rotated/nms_rotated.egg-info/SOURCES.txt
@@ -0,0 +1,10 @@
+setup.py
+nms_rotated.egg-info/PKG-INFO
+nms_rotated.egg-info/SOURCES.txt
+nms_rotated.egg-info/dependency_links.txt
+nms_rotated.egg-info/not-zip-safe
+nms_rotated.egg-info/top_level.txt
+src/nms_rotated_cpu.cpp
+src/nms_rotated_cuda.cu
+src/nms_rotated_ext.cpp
+src/poly_nms_cuda.cu
\ No newline at end of file
diff --git a/utils/nms_rotated/nms_rotated.egg-info/dependency_links.txt b/utils/nms_rotated/nms_rotated.egg-info/dependency_links.txt
new file mode 100644
index 0000000..8b13789
--- /dev/null
+++ b/utils/nms_rotated/nms_rotated.egg-info/dependency_links.txt
@@ -0,0 +1 @@
+
diff --git a/utils/nms_rotated/nms_rotated.egg-info/not-zip-safe b/utils/nms_rotated/nms_rotated.egg-info/not-zip-safe
new file mode 100644
index 0000000..8b13789
--- /dev/null
+++ b/utils/nms_rotated/nms_rotated.egg-info/not-zip-safe
@@ -0,0 +1 @@
+
diff --git a/utils/nms_rotated/nms_rotated.egg-info/top_level.txt b/utils/nms_rotated/nms_rotated.egg-info/top_level.txt
new file mode 100644
index 0000000..8b13789
--- /dev/null
+++ b/utils/nms_rotated/nms_rotated.egg-info/top_level.txt
@@ -0,0 +1 @@
+
diff --git a/utils/nms_rotated/nms_rotated_ext.pyd b/utils/nms_rotated/nms_rotated_ext.pyd
new file mode 100644
index 0000000..ec6c224
Binary files /dev/null and b/utils/nms_rotated/nms_rotated_ext.pyd differ
diff --git a/utils/nms_rotated/nms_rotated_wrapper.py b/utils/nms_rotated/nms_rotated_wrapper.py
new file mode 100644
index 0000000..b2412e5
--- /dev/null
+++ b/utils/nms_rotated/nms_rotated_wrapper.py
@@ -0,0 +1,79 @@
+import numpy as np
+import torch
+
+from . import nms_rotated_ext
+
+
+def obb_nms(dets, scores, iou_thr, device_id=None):
+ """
+ RIoU NMS - iou_thr.
+ Args:
+ dets (tensor/array): (num, [cx cy w h θ]) θ∈[-pi/2, pi/2)
+ scores (tensor/array): (num)
+ iou_thr (float): (1)
+ Returns:
+ dets (tensor): (n_nms, [cx cy w h θ])
+ inds (tensor): (n_nms), nms index of dets
+ """
+ if isinstance(dets, torch.Tensor):
+ is_numpy = False
+ dets_th = dets
+ elif isinstance(dets, np.ndarray):
+ is_numpy = True
+ device = 'cpu' if device_id is None else f'cuda:{device_id}'
+ dets_th = torch.from_numpy(dets).to(device)
+ else:
+ raise TypeError('dets must be eithr a Tensor or numpy array, '
+ f'but got {type(dets)}')
+
+ if dets_th.numel() == 0: # len(dets)
+ inds = dets_th.new_zeros(0, dtype=torch.int64)
+ else:
+ # same bug will happen when bboxes is too small
+ too_small = dets_th[:, [2, 3]].min(1)[0] < 0.001 # [n]
+ if too_small.all(): # all the bboxes is too small
+ inds = dets_th.new_zeros(0, dtype=torch.int64)
+ else:
+ ori_inds = torch.arange(dets_th.size(0)) # 0 ~ n-1
+ ori_inds = ori_inds[~too_small]
+ dets_th = dets_th[~too_small] # (n_filter, 5)
+ scores = scores[~too_small]
+
+ inds = nms_rotated_ext.nms_rotated(dets_th, scores, iou_thr)
+ inds = ori_inds[inds]
+
+ if is_numpy:
+ inds = inds.cpu().numpy()
+ return dets[inds, :], inds
+
+
+def poly_nms(dets, iou_thr, device_id=None):
+ if isinstance(dets, torch.Tensor):
+ is_numpy = False
+ dets_th = dets
+ elif isinstance(dets, np.ndarray):
+ is_numpy = True
+ device = 'cpu' if device_id is None else f'cuda:{device_id}'
+ dets_th = torch.from_numpy(dets).to(device)
+ else:
+ raise TypeError('dets must be eithr a Tensor or numpy array, '
+ f'but got {type(dets)}')
+
+ if dets_th.device == torch.device('cpu'):
+ raise NotImplementedError
+ inds = nms_rotated_ext.nms_poly(dets_th.float(), iou_thr)
+
+ if is_numpy:
+ inds = inds.cpu().numpy()
+ return dets[inds, :], inds
+
+if __name__ == '__main__':
+ rboxes_opencv = torch.tensor(([136.6, 111.6, 200, 100, -60],
+ [136.6, 111.6, 100, 200, -30],
+ [100, 100, 141.4, 141.4, -45],
+ [100, 100, 141.4, 141.4, -45]))
+ rboxes_longedge = torch.tensor(([136.6, 111.6, 200, 100, -60],
+ [136.6, 111.6, 200, 100, 120],
+ [100, 100, 141.4, 141.4, 45],
+ [100, 100, 141.4, 141.4, 135]))
+
\ No newline at end of file
diff --git a/utils/nms_rotated/setup.py b/utils/nms_rotated/setup.py
new file mode 100644
index 0000000..a3ee967
--- /dev/null
+++ b/utils/nms_rotated/setup.py
@@ -0,0 +1,54 @@
+#!/usr/bin/env python
+import os
+import subprocess
+import time
+from setuptools import find_packages, setup
+
+import torch
+from torch.utils.cpp_extension import (BuildExtension, CppExtension,
+ CUDAExtension)
+def make_cuda_ext(name, module, sources, sources_cuda=[]):
+
+ define_macros = []
+ extra_compile_args = {'cxx': []}
+
+ if torch.cuda.is_available() or os.getenv('FORCE_CUDA', '0') == '1':
+ define_macros += [('WITH_CUDA', None)]
+ extension = CUDAExtension
+ extra_compile_args['nvcc'] = [
+ '-D__CUDA_NO_HALF_OPERATORS__',
+ '-D__CUDA_NO_HALF_CONVERSIONS__',
+ '-D__CUDA_NO_HALF2_OPERATORS__',
+ ]
+ sources += sources_cuda
+ else:
+ print(f'Compiling {name} without CUDA')
+ extension = CppExtension
+ # raise EnvironmentError('CUDA is required to compile MMDetection!')
+
+ return extension(
+ name=f'{module}.{name}',
+ sources=[os.path.join(*module.split('.'), p) for p in sources],
+ define_macros=define_macros,
+ extra_compile_args=extra_compile_args)
+
+# python setup.py develop
+if __name__ == '__main__':
+ #write_version_py()
+ setup(
+ name='nms_rotated',
+ ext_modules=[
+ make_cuda_ext(
+ name='nms_rotated_ext',
+ module='',
+ sources=[
+ 'src/nms_rotated_cpu.cpp',
+ 'src/nms_rotated_ext.cpp'
+ ],
+ sources_cuda=[
+ 'src/nms_rotated_cuda.cu',
+ 'src/poly_nms_cuda.cu',
+ ]),
+ ],
+ cmdclass={'build_ext': BuildExtension},
+ zip_safe=False)
\ No newline at end of file
diff --git a/utils/nms_rotated/src/box_iou_rotated_utils.h b/utils/nms_rotated/src/box_iou_rotated_utils.h
new file mode 100644
index 0000000..c017e17
--- /dev/null
+++ b/utils/nms_rotated/src/box_iou_rotated_utils.h
@@ -0,0 +1,360 @@
+// Mortified from
+// https://github.com/facebookresearch/detectron2/tree/master/detectron2/layers/csrc/box_iou_rotated
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+#pragma once
+
+#include
+#include
+
+#if defined(__CUDACC__) || __HCC__ == 1 || __HIP__ == 1
+// Designates functions callable from the host (CPU) and the device (GPU)
+#define HOST_DEVICE __host__ __device__
+#define HOST_DEVICE_INLINE HOST_DEVICE __forceinline__
+#else
+#include
+#define HOST_DEVICE
+#define HOST_DEVICE_INLINE HOST_DEVICE inline
+#endif
+
+
+template
+struct RotatedBox {
+ T x_ctr, y_ctr, w, h, a;
+};
+
+template
+struct Point {
+ T x, y;
+ HOST_DEVICE_INLINE Point(const T& px = 0, const T& py = 0) : x(px), y(py) {}
+ HOST_DEVICE_INLINE Point operator+(const Point& p) const {
+ return Point(x + p.x, y + p.y);
+ }
+ HOST_DEVICE_INLINE Point& operator+=(const Point& p) {
+ x += p.x;
+ y += p.y;
+ return *this;
+ }
+ HOST_DEVICE_INLINE Point operator-(const Point& p) const {
+ return Point(x - p.x, y - p.y);
+ }
+ HOST_DEVICE_INLINE Point operator*(const T coeff) const {
+ return Point(x * coeff, y * coeff);
+ }
+};
+
+template
+HOST_DEVICE_INLINE T dot_2d(const Point& A, const Point& B) {
+ return A.x * B.x + A.y * B.y;
+}
+
+// R: result type. can be different from input type
+template
+HOST_DEVICE_INLINE R cross_2d(const Point& A, const Point& B) {
+ return static_cast(A.x) * static_cast(B.y) -
+ static_cast(B.x) * static_cast(A.y);
+}
+
+template
+HOST_DEVICE_INLINE void get_rotated_vertices(
+ const RotatedBox& box,
+ Point (&pts)[4]) {
+ // M_PI / 180. == 0.01745329251
+ //double theta = box.a * 0.01745329251; ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ double theta = box.a;
+ T cosTheta2 = (T)cos(theta) * 0.5f;
+ T sinTheta2 = (T)sin(theta) * 0.5f;
+
+ // y: top --> down; x: left --> right
+ pts[0].x = box.x_ctr + sinTheta2 * box.h + cosTheta2 * box.w;
+ pts[0].y = box.y_ctr + cosTheta2 * box.h - sinTheta2 * box.w;
+ pts[1].x = box.x_ctr - sinTheta2 * box.h + cosTheta2 * box.w;
+ pts[1].y = box.y_ctr - cosTheta2 * box.h - sinTheta2 * box.w;
+ pts[2].x = 2 * box.x_ctr - pts[0].x;
+ pts[2].y = 2 * box.y_ctr - pts[0].y;
+ pts[3].x = 2 * box.x_ctr - pts[1].x;
+ pts[3].y = 2 * box.y_ctr - pts[1].y;
+}
+
+template
+HOST_DEVICE_INLINE int get_intersection_points(
+ const Point (&pts1)[4],
+ const Point (&pts2)[4],
+ Point (&intersections)[24]) {
+ // Line vector
+ // A line from p1 to p2 is: p1 + (p2-p1)*t, t=[0,1]
+ Point vec1[4], vec2[4];
+ for (int i = 0; i < 4; i++) {
+ vec1[i] = pts1[(i + 1) % 4] - pts1[i];
+ vec2[i] = pts2[(i + 1) % 4] - pts2[i];
+ }
+
+ // Line test - test all line combos for intersection
+ int num = 0; // number of intersections
+ for (int i = 0; i < 4; i++) {
+ for (int j = 0; j < 4; j++) {
+ // Solve for 2x2 Ax=b
+ T det = cross_2d(vec2[j], vec1[i]);
+
+ // This takes care of parallel lines
+ if (fabs(det) <= 1e-14) {
+ continue;
+ }
+
+ auto vec12 = pts2[j] - pts1[i];
+
+ T t1 = cross_2d(vec2[j], vec12) / det;
+ T t2 = cross_2d(vec1[i], vec12) / det;
+
+ if (t1 >= 0.0f && t1 <= 1.0f && t2 >= 0.0f && t2 <= 1.0f) {
+ intersections[num++] = pts1[i] + vec1[i] * t1;
+ }
+ }
+ }
+
+ // Check for vertices of rect1 inside rect2
+ {
+ const auto& AB = vec2[0];
+ const auto& DA = vec2[3];
+ auto ABdotAB = dot_2d(AB, AB);
+ auto ADdotAD = dot_2d(DA, DA);
+ for (int i = 0; i < 4; i++) {
+ // assume ABCD is the rectangle, and P is the point to be judged
+ // P is inside ABCD iff. P's projection on AB lies within AB
+ // and P's projection on AD lies within AD
+
+ auto AP = pts1[i] - pts2[0];
+
+ auto APdotAB = dot_2d(AP, AB);
+ auto APdotAD = -dot_2d(AP, DA);
+
+ if ((APdotAB >= 0) && (APdotAD >= 0) && (APdotAB <= ABdotAB) &&
+ (APdotAD <= ADdotAD)) {
+ intersections[num++] = pts1[i];
+ }
+ }
+ }
+
+ // Reverse the check - check for vertices of rect2 inside rect1
+ {
+ const auto& AB = vec1[0];
+ const auto& DA = vec1[3];
+ auto ABdotAB = dot_2d(AB, AB);
+ auto ADdotAD = dot_2d(DA, DA);
+ for (int i = 0; i < 4; i++) {
+ auto AP = pts2[i] - pts1[0];
+
+ auto APdotAB = dot_2d(AP, AB);
+ auto APdotAD = -dot_2d(AP, DA);
+
+ if ((APdotAB >= 0) && (APdotAD >= 0) && (APdotAB <= ABdotAB) &&
+ (APdotAD <= ADdotAD)) {
+ intersections[num++] = pts2[i];
+ }
+ }
+ }
+
+ return num;
+}
+
+template
+HOST_DEVICE_INLINE int convex_hull_graham(
+ const Point (&p)[24],
+ const int& num_in,
+ Point (&q)[24],
+ bool shift_to_zero = false) {
+ assert(num_in >= 2);
+
+ // Step 1:
+ // Find point with minimum y
+ // if more than 1 points have the same minimum y,
+ // pick the one with the minimum x.
+ int t = 0;
+ for (int i = 1; i < num_in; i++) {
+ if (p[i].y < p[t].y || (p[i].y == p[t].y && p[i].x < p[t].x)) {
+ t = i;
+ }
+ }
+ auto& start = p[t]; // starting point
+
+ // Step 2:
+ // Subtract starting point from every points (for sorting in the next step)
+ for (int i = 0; i < num_in; i++) {
+ q[i] = p[i] - start;
+ }
+
+ // Swap the starting point to position 0
+ auto tmp = q[0];
+ q[0] = q[t];
+ q[t] = tmp;
+
+ // Step 3:
+ // Sort point 1 ~ num_in according to their relative cross-product values
+ // (essentially sorting according to angles)
+ // If the angles are the same, sort according to their distance to origin
+ T dist[24];
+#if defined(__CUDACC__) || __HCC__ == 1 || __HIP__ == 1
+ // compute distance to origin before sort, and sort them together with the
+ // points
+ for (int i = 0; i < num_in; i++) {
+ dist[i] = dot_2d(q[i], q[i]);
+ }
+
+ // CUDA version
+ // In the future, we can potentially use thrust
+ // for sorting here to improve speed (though not guaranteed)
+ for (int i = 1; i < num_in - 1; i++) {
+ for (int j = i + 1; j < num_in; j++) {
+ T crossProduct = cross_2d(q[i], q[j]);
+ if ((crossProduct < -1e-6) ||
+ (fabs(crossProduct) < 1e-6 && dist[i] > dist[j])) {
+ auto q_tmp = q[i];
+ q[i] = q[j];
+ q[j] = q_tmp;
+ auto dist_tmp = dist[i];
+ dist[i] = dist[j];
+ dist[j] = dist_tmp;
+ }
+ }
+ }
+#else
+ // CPU version
+ std::sort(
+ q + 1, q + num_in, [](const Point& A, const Point& B) -> bool {
+ T temp = cross_2d(A, B);
+ if (fabs(temp) < 1e-6) {
+ return dot_2d(A, A) < dot_2d(B, B);
+ } else {
+ return temp > 0;
+ }
+ });
+ // compute distance to origin after sort, since the points are now different.
+ for (int i = 0; i < num_in; i++) {
+ dist[i] = dot_2d(q[i], q[i]);
+ }
+#endif
+
+ // Step 4:
+ // Make sure there are at least 2 points (that don't overlap with each other)
+ // in the stack
+ int k; // index of the non-overlapped second point
+ for (k = 1; k < num_in; k++) {
+ if (dist[k] > 1e-8) {
+ break;
+ }
+ }
+ if (k == num_in) {
+ // We reach the end, which means the convex hull is just one point
+ q[0] = p[t];
+ return 1;
+ }
+ q[1] = q[k];
+ int m = 2; // 2 points in the stack
+ // Step 5:
+ // Finally we can start the scanning process.
+ // When a non-convex relationship between the 3 points is found
+ // (either concave shape or duplicated points),
+ // we pop the previous point from the stack
+ // until the 3-point relationship is convex again, or
+ // until the stack only contains two points
+ for (int i = k + 1; i < num_in; i++) {
+ while (m > 1) {
+ auto q1 = q[i] - q[m - 2], q2 = q[m - 1] - q[m - 2];
+ // cross_2d() uses FMA and therefore computes round(round(q1.x*q2.y) -
+ // q2.x*q1.y) So it may not return 0 even when q1==q2. Therefore we
+ // compare round(q1.x*q2.y) and round(q2.x*q1.y) directly. (round means
+ // round to nearest floating point).
+ if (q1.x * q2.y >= q2.x * q1.y)
+ m--;
+ else
+ break;
+ }
+ // Using double also helps, but float can solve the issue for now.
+ // while (m > 1 && cross_2d(q[i] - q[m - 2], q[m - 1] - q[m - 2])
+ // >= 0) {
+ // m--;
+ // }
+ q[m++] = q[i];
+ }
+
+ // Step 6 (Optional):
+ // In general sense we need the original coordinates, so we
+ // need to shift the points back (reverting Step 2)
+ // But if we're only interested in getting the area/perimeter of the shape
+ // We can simply return.
+ if (!shift_to_zero) {
+ for (int i = 0; i < m; i++) {
+ q[i] += start;
+ }
+ }
+
+ return m;
+}
+
+template
+HOST_DEVICE_INLINE T polygon_area(const Point (&q)[24], const int& m) {
+ if (m <= 2) {
+ return 0;
+ }
+
+ T area = 0;
+ for (int i = 1; i < m - 1; i++) {
+ area += fabs(cross_2d(q[i] - q[0], q[i + 1] - q[0]));
+ }
+
+ return area / 2.0;
+}
+
+template
+HOST_DEVICE_INLINE T rotated_boxes_intersection(
+ const RotatedBox& box1,
+ const RotatedBox& box2) {
+ // There are up to 4 x 4 + 4 + 4 = 24 intersections (including dups) returned
+ // from rotated_rect_intersection_pts
+ Point intersectPts[24], orderedPts[24];
+
+ Point pts1[4];
+ Point pts2[4];
+ get_rotated_vertices(box1, pts1);
+ get_rotated_vertices(box2, pts2);
+
+ int num = get_intersection_points(pts1, pts2, intersectPts);
+
+ if (num <= 2) {
+ return 0.0;
+ }
+
+ // Convex Hull to order the intersection points in clockwise order and find
+ // the contour area.
+ int num_convex = convex_hull_graham(intersectPts, num, orderedPts, true);
+ return polygon_area(orderedPts, num_convex);
+}
+
+
+template
+HOST_DEVICE_INLINE T
+single_box_iou_rotated(T const* const box1_raw, T const* const box2_raw) {
+ // shift center to the middle point to achieve higher precision in result
+ RotatedBox box1, box2;
+ auto center_shift_x = (box1_raw[0] + box2_raw[0]) / 2.0;
+ auto center_shift_y = (box1_raw[1] + box2_raw[1]) / 2.0;
+ box1.x_ctr = box1_raw[0] - center_shift_x;
+ box1.y_ctr = box1_raw[1] - center_shift_y;
+ box1.w = box1_raw[2];
+ box1.h = box1_raw[3];
+ box1.a = box1_raw[4];
+ box2.x_ctr = box2_raw[0] - center_shift_x;
+ box2.y_ctr = box2_raw[1] - center_shift_y;
+ box2.w = box2_raw[2];
+ box2.h = box2_raw[3];
+ box2.a = box2_raw[4];
+
+ T area1 = box1.w * box1.h;
+ T area2 = box2.w * box2.h;
+ if (area1 < 1e-14 || area2 < 1e-14) {
+ return 0.f;
+ }
+
+ T intersection = rotated_boxes_intersection(box1, box2);
+ T iou = intersection / (area1 + area2 - intersection);
+ return iou;
+}
diff --git a/utils/nms_rotated/src/nms_rotated_cpu.cpp b/utils/nms_rotated/src/nms_rotated_cpu.cpp
new file mode 100644
index 0000000..185e9a4
--- /dev/null
+++ b/utils/nms_rotated/src/nms_rotated_cpu.cpp
@@ -0,0 +1,74 @@
+// Modified from
+// https://github.com/facebookresearch/detectron2/tree/master/detectron2/layers/csrc/nms_rotated
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+#include
+#include "box_iou_rotated_utils.h"
+
+
+template
+at::Tensor nms_rotated_cpu_kernel(
+ const at::Tensor& dets,
+ const at::Tensor& scores,
+ const float iou_threshold) {
+ // nms_rotated_cpu_kernel is modified from torchvision's nms_cpu_kernel,
+ // however, the code in this function is much shorter because
+ // we delegate the IoU computation for rotated boxes to
+ // the single_box_iou_rotated function in box_iou_rotated_utils.h
+ AT_ASSERTM(dets.device().is_cpu(), "dets must be a CPU tensor");
+ AT_ASSERTM(scores.device().is_cpu(), "scores must be a CPU tensor");
+ AT_ASSERTM(
+ dets.scalar_type() == scores.scalar_type(),
+ "dets should have the same type as scores");
+
+ if (dets.numel() == 0) {
+ return at::empty({0}, dets.options().dtype(at::kLong));
+ }
+
+ auto order_t = std::get<1>(scores.sort(0, /* descending=*/true));
+
+ auto ndets = dets.size(0);
+ at::Tensor suppressed_t = at::zeros({ndets}, dets.options().dtype(at::kByte));
+ at::Tensor keep_t = at::zeros({ndets}, dets.options().dtype(at::kLong));
+
+ auto suppressed = suppressed_t.data_ptr();
+ auto keep = keep_t.data_ptr();
+ auto order = order_t.data_ptr();
+
+ int64_t num_to_keep = 0;
+
+ for (int64_t _i = 0; _i < ndets; _i++) {
+ auto i = order[_i];
+ if (suppressed[i] == 1) {
+ continue;
+ }
+
+ keep[num_to_keep++] = i;
+
+ for (int64_t _j = _i + 1; _j < ndets; _j++) {
+ auto j = order[_j];
+ if (suppressed[j] == 1) {
+ continue;
+ }
+
+ auto ovr = single_box_iou_rotated(
+ dets[i].data_ptr(), dets[j].data_ptr());
+ if (ovr >= iou_threshold) {
+ suppressed[j] = 1;
+ }
+ }
+ }
+ return keep_t.narrow(/*dim=*/0, /*start=*/0, /*length=*/num_to_keep);
+}
+
+at::Tensor nms_rotated_cpu(
+ // input must be contiguous
+ const at::Tensor& dets,
+ const at::Tensor& scores,
+ const float iou_threshold) {
+ auto result = at::empty({0}, dets.options());
+
+ AT_DISPATCH_FLOATING_TYPES(dets.scalar_type(), "nms_rotated", [&] {
+ result = nms_rotated_cpu_kernel(dets, scores, iou_threshold);
+ });
+ return result;
+}
diff --git a/utils/nms_rotated/src/nms_rotated_cuda.cu b/utils/nms_rotated/src/nms_rotated_cuda.cu
new file mode 100644
index 0000000..84d5acf
--- /dev/null
+++ b/utils/nms_rotated/src/nms_rotated_cuda.cu
@@ -0,0 +1,134 @@
+// Modified from
+// https://github.com/facebookresearch/detectron2/tree/master/detectron2/layers/csrc/nms_rotated
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+#include
+#include
+#include
+#include
+#include "box_iou_rotated_utils.h"
+
+int const threadsPerBlock = sizeof(unsigned long long) * 8;
+
+template
+__global__ void nms_rotated_cuda_kernel(
+ const int n_boxes,
+ const float iou_threshold,
+ const T* dev_boxes,
+ unsigned long long* dev_mask) {
+ // nms_rotated_cuda_kernel is modified from torchvision's nms_cuda_kernel
+
+ const int row_start = blockIdx.y;
+ const int col_start = blockIdx.x;
+
+ // if (row_start > col_start) return;
+
+ const int row_size =
+ min(n_boxes - row_start * threadsPerBlock, threadsPerBlock);
+ const int col_size =
+ min(n_boxes - col_start * threadsPerBlock, threadsPerBlock);
+
+ // Compared to nms_cuda_kernel, where each box is represented with 4 values
+ // (x1, y1, x2, y2), each rotated box is represented with 5 values
+ // (x_center, y_center, width, height, angle_degrees) here.
+ __shared__ T block_boxes[threadsPerBlock * 5];
+ if (threadIdx.x < col_size) {
+ block_boxes[threadIdx.x * 5 + 0] =
+ dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 0];
+ block_boxes[threadIdx.x * 5 + 1] =
+ dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 1];
+ block_boxes[threadIdx.x * 5 + 2] =
+ dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 2];
+ block_boxes[threadIdx.x * 5 + 3] =
+ dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 3];
+ block_boxes[threadIdx.x * 5 + 4] =
+ dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 4];
+ }
+ __syncthreads();
+
+ if (threadIdx.x < row_size) {
+ const int cur_box_idx = threadsPerBlock * row_start + threadIdx.x;
+ const T* cur_box = dev_boxes + cur_box_idx * 5;
+ int i = 0;
+ unsigned long long t = 0;
+ int start = 0;
+ if (row_start == col_start) {
+ start = threadIdx.x + 1;
+ }
+ for (i = start; i < col_size; i++) {
+ // Instead of devIoU used by original horizontal nms, here
+ // we use the single_box_iou_rotated function from box_iou_rotated_utils.h
+ if (single_box_iou_rotated(cur_box, block_boxes + i * 5) >
+ iou_threshold) {
+ t |= 1ULL << i;
+ }
+ }
+ const int col_blocks = at::cuda::ATenCeilDiv(n_boxes, threadsPerBlock);
+ dev_mask[cur_box_idx * col_blocks + col_start] = t;
+ }
+}
+
+
+at::Tensor nms_rotated_cuda(
+ // input must be contiguous
+ const at::Tensor& dets,
+ const at::Tensor& scores,
+ float iou_threshold) {
+ // using scalar_t = float;
+ AT_ASSERTM(dets.is_cuda(), "dets must be a CUDA tensor");
+ AT_ASSERTM(scores.is_cuda(), "scores must be a CUDA tensor");
+ at::cuda::CUDAGuard device_guard(dets.device());
+
+ auto order_t = std::get<1>(scores.sort(0, /* descending=*/true));
+ auto dets_sorted = dets.index_select(0, order_t);
+
+ auto dets_num = dets.size(0);
+
+ const int col_blocks =
+ at::cuda::ATenCeilDiv(static_cast(dets_num), threadsPerBlock);
+
+ at::Tensor mask =
+ at::empty({dets_num * col_blocks}, dets.options().dtype(at::kLong));
+
+ dim3 blocks(col_blocks, col_blocks);
+ dim3 threads(threadsPerBlock);
+ cudaStream_t stream = at::cuda::getCurrentCUDAStream();
+
+ AT_DISPATCH_FLOATING_TYPES(
+ dets_sorted.scalar_type(), "nms_rotated_kernel_cuda", [&] {
+ nms_rotated_cuda_kernel<<>>(
+ dets_num,
+ iou_threshold,
+ dets_sorted.data_ptr(),
+ (unsigned long long*)mask.data_ptr());
+ });
+
+ at::Tensor mask_cpu = mask.to(at::kCPU);
+ unsigned long long* mask_host =
+ (unsigned long long*)mask_cpu.data_ptr();
+
+ std::vector remv(col_blocks);
+ memset(&remv[0], 0, sizeof(unsigned long long) * col_blocks);
+
+ at::Tensor keep =
+ at::empty({dets_num}, dets.options().dtype(at::kLong).device(at::kCPU));
+ int64_t* keep_out = keep.data_ptr();
+
+ int num_to_keep = 0;
+ for (int i = 0; i < dets_num; i++) {
+ int nblock = i / threadsPerBlock;
+ int inblock = i % threadsPerBlock;
+
+ if (!(remv[nblock] & (1ULL << inblock))) {
+ keep_out[num_to_keep++] = i;
+ unsigned long long* p = mask_host + i * col_blocks;
+ for (int j = nblock; j < col_blocks; j++) {
+ remv[j] |= p[j];
+ }
+ }
+ }
+
+ AT_CUDA_CHECK(cudaGetLastError());
+ return order_t.index(
+ {keep.narrow(/*dim=*/0, /*start=*/0, /*length=*/num_to_keep)
+ .to(order_t.device(), keep.scalar_type())});
+}
diff --git a/utils/nms_rotated/src/nms_rotated_ext.cpp b/utils/nms_rotated/src/nms_rotated_ext.cpp
new file mode 100644
index 0000000..287338f
--- /dev/null
+++ b/utils/nms_rotated/src/nms_rotated_ext.cpp
@@ -0,0 +1,60 @@
+// Modified from
+// https://github.com/facebookresearch/detectron2/tree/master/detectron2/layers/csrc/nms_rotated
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+#include
+#include
+
+
+#ifdef WITH_CUDA
+at::Tensor nms_rotated_cuda(
+ const at::Tensor& dets,
+ const at::Tensor& scores,
+ const float iou_threshold);
+
+at::Tensor poly_nms_cuda(
+ const at::Tensor boxes,
+ float nms_overlap_thresh);
+#endif
+
+at::Tensor nms_rotated_cpu(
+ const at::Tensor& dets,
+ const at::Tensor& scores,
+ const float iou_threshold);
+
+
+inline at::Tensor nms_rotated(
+ const at::Tensor& dets,
+ const at::Tensor& scores,
+ const float iou_threshold) {
+ assert(dets.device().is_cuda() == scores.device().is_cuda());
+ if (dets.device().is_cuda()) {
+#ifdef WITH_CUDA
+ return nms_rotated_cuda(
+ dets.contiguous(), scores.contiguous(), iou_threshold);
+#else
+ AT_ERROR("Not compiled with GPU support");
+#endif
+ }
+ return nms_rotated_cpu(dets.contiguous(), scores.contiguous(), iou_threshold);
+}
+
+
+inline at::Tensor nms_poly(
+ const at::Tensor& dets,
+ const float iou_threshold) {
+ if (dets.device().is_cuda()) {
+#ifdef WITH_CUDA
+ if (dets.numel() == 0)
+ return at::empty({0}, dets.options().dtype(at::kLong).device(at::kCPU));
+ return poly_nms_cuda(dets, iou_threshold);
+#else
+ AT_ERROR("POLY_NMS is not compiled with GPU support");
+#endif
+ }
+ AT_ERROR("POLY_NMS is not implemented on CPU");
+}
+
+PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
+ m.def("nms_rotated", &nms_rotated, "nms for rotated bboxes");
+ m.def("nms_poly", &nms_poly, "nms for poly bboxes");
+}
diff --git a/utils/nms_rotated/src/poly_nms_cpu.cpp b/utils/nms_rotated/src/poly_nms_cpu.cpp
new file mode 100644
index 0000000..75af948
--- /dev/null
+++ b/utils/nms_rotated/src/poly_nms_cpu.cpp
@@ -0,0 +1,5 @@
+#include
+
+template
+at::Tensor poly_nms_cpu_kernel(const at::Tensor& dets, const float threshold) {
+
diff --git a/utils/nms_rotated/src/poly_nms_cuda.cu b/utils/nms_rotated/src/poly_nms_cuda.cu
new file mode 100644
index 0000000..9fc017d
--- /dev/null
+++ b/utils/nms_rotated/src/poly_nms_cuda.cu
@@ -0,0 +1,255 @@
+#include
+#include
+
+#include
+#include
+
+#include
+#include
+
+#define CUDA_CHECK(condition) \
+ /* Code block avoids redefinition of cudaError_t error */ \
+ do { \
+ cudaError_t error = condition; \
+ if (error != cudaSuccess) { \
+ std::cout << cudaGetErrorString(error) << std::endl; \
+ } \
+ } while (0)
+
+#define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0))
+int const threadsPerBlock = sizeof(unsigned long long) * 8;
+
+
+ // #define maxn 10
+// const double eps=1E-8;
+//
+// __device__ inline int sig(float d){
+// return(d>eps)-(d<-eps);
+// }
+#define maxn 10
+//const double eps = 0.00000001;
+
+__device__ inline int sig(float d) {
+ return (d > 0.00000001) - (d < -0.00000001);
+}
+
+__device__ inline int point_eq(const float2 a, const float2 b) {
+ return sig(a.x - b.x) == 0 && sig(a.y - b.y) == 0;
+}
+
+__device__ inline void point_swap(float2 *a, float2 *b) {
+ float2 temp = *a;
+ *a = *b;
+ *b = temp;
+}
+
+__device__ inline void point_reverse(float2 *first, float2* last)
+{
+ while ((first != last) && (first != --last)) {
+ point_swap(first, last);
+ ++first;
+ }
+}
+
+__device__ inline float cross(float2 o, float2 a, float2 b) { //叉积
+ return(a.x - o.x)*(b.y - o.y) - (b.x - o.x)*(a.y - o.y);
+}
+__device__ inline float area(float2* ps, int n) {
+ ps[n] = ps[0];
+ float res = 0;
+ for (int i = 0; i < n; i++) {
+ res += ps[i].x*ps[i + 1].y - ps[i].y*ps[i + 1].x;
+ }
+ return res / 2.0;
+}
+__device__ inline int lineCross(float2 a, float2 b, float2 c, float2 d, float2&p) {
+ float s1, s2;
+ s1 = cross(a, b, c);
+ s2 = cross(a, b, d);
+ if (sig(s1) == 0 && sig(s2) == 0) return 2;
+ if (sig(s2 - s1) == 0) return 0;
+ p.x = (c.x*s2 - d.x*s1) / (s2 - s1);
+ p.y = (c.y*s2 - d.y*s1) / (s2 - s1);
+ return 1;
+}
+
+__device__ inline void polygon_cut(float2*p, int&n, float2 a, float2 b, float2* pp) {
+
+ int m = 0; p[n] = p[0];
+ for (int i = 0; i < n; i++) {
+ if (sig(cross(a, b, p[i])) > 0) pp[m++] = p[i];
+ if (sig(cross(a, b, p[i])) != sig(cross(a, b, p[i + 1])))
+ lineCross(a, b, p[i], p[i + 1], pp[m++]);
+ }
+ n = 0;
+ for (int i = 0; i < m; i++)
+ if (!i || !(point_eq(pp[i], pp[i - 1])))
+ p[n++] = pp[i];
+ // while(n>1&&p[n-1]==p[0])n--;
+ while (n > 1 && point_eq(p[n - 1], p[0]))n--;
+}
+
+//---------------华丽的分隔线-----------------//
+//返回三角形oab和三角形ocd的有向交面积,o是原点//
+__device__ inline float intersectArea(float2 a, float2 b, float2 c, float2 d) {
+ float2 o = make_float2(0, 0);
+ int s1 = sig(cross(o, a, b));
+ int s2 = sig(cross(o, c, d));
+ if (s1 == 0 || s2 == 0)return 0.0;//退化,面积为0
+ // if(s1==-1) swap(a,b);
+ // if(s2==-1) swap(c,d);
+ if (s1 == -1) point_swap(&a, &b);
+ if (s2 == -1) point_swap(&c, &d);
+ float2 p[10] = { o,a,b };
+ int n = 3;
+ float2 pp[maxn];
+ polygon_cut(p, n, o, c, pp);
+ polygon_cut(p, n, c, d, pp);
+ polygon_cut(p, n, d, o, pp);
+ float res = fabs(area(p, n));
+ if (s1*s2 == -1) res = -res; return res;
+}
+
+//求两多边形的交面积
+__device__ inline float intersectArea(float2 * ps1, int n1, float2 * ps2, int n2) {
+ if (area(ps1, n1) < 0) point_reverse(ps1, ps1 + n1);
+ if (area(ps2, n2) < 0) point_reverse(ps2, ps2 + n2);
+ ps1[n1] = ps1[0];
+ ps2[n2] = ps2[0];
+ float res = 0;
+ for (int i = 0; i < n1; i++) {
+ for (int j = 0; j < n2; j++) {
+ res += intersectArea(ps1[i], ps1[i + 1], ps2[j], ps2[j + 1]);
+ }
+ }
+ return res;//assumeresispositive!
+}
+
+// TODO: optimal if by first calculate the iou between two hbbs
+__device__ inline float devPolyIoU(float const * const p, float const * const q) {
+ float2 ps1[maxn], ps2[maxn];
+ int n1 = 4;
+ int n2 = 4;
+ for (int i = 0; i < 4; i++) {
+ ps1[i].x = p[i * 2];
+ ps1[i].y = p[i * 2 + 1];
+
+ ps2[i].x = q[i * 2];
+ ps2[i].y = q[i * 2 + 1];
+ }
+ float inter_area = intersectArea(ps1, n1, ps2, n2);
+ float union_area = fabs(area(ps1, n1)) + fabs(area(ps2, n2)) - inter_area;
+ float iou = 0;
+ if (union_area == 0) {
+ iou = (inter_area + 1) / (union_area + 1);
+ }
+ else {
+ iou = inter_area / union_area;
+ }
+ return iou;
+}
+
+__global__ void poly_nms_kernel(const int n_polys, const float nms_overlap_thresh,
+ const float *dev_polys, unsigned long long *dev_mask) {
+ const int row_start = blockIdx.y;
+ const int col_start = blockIdx.x;
+
+ const int row_size =
+ min(n_polys - row_start * threadsPerBlock, threadsPerBlock);
+ const int cols_size =
+ min(n_polys - col_start * threadsPerBlock, threadsPerBlock);
+
+ __shared__ float block_polys[threadsPerBlock * 9];
+ if (threadIdx.x < cols_size) {
+ block_polys[threadIdx.x * 9 + 0] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 0];
+ block_polys[threadIdx.x * 9 + 1] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 1];
+ block_polys[threadIdx.x * 9 + 2] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 2];
+ block_polys[threadIdx.x * 9 + 3] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 3];
+ block_polys[threadIdx.x * 9 + 4] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 4];
+ block_polys[threadIdx.x * 9 + 5] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 5];
+ block_polys[threadIdx.x * 9 + 6] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 6];
+ block_polys[threadIdx.x * 9 + 7] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 7];
+ block_polys[threadIdx.x * 9 + 8] = dev_polys[(threadsPerBlock * col_start + threadIdx.x) * 9 + 8];
+ }
+ __syncthreads();
+ if (threadIdx.x < row_size) {
+ const int cur_box_idx = threadsPerBlock * row_start + threadIdx.x;
+ const float *cur_box = dev_polys + cur_box_idx * 9;
+ int i = 0;
+ unsigned long long t = 0;
+ int start = 0;
+ if (row_start == col_start) {
+ start = threadIdx.x + 1;
+ }
+ for (i = start; i < cols_size; i++) {
+ if (devPolyIoU(cur_box, block_polys + i * 9) > nms_overlap_thresh) {
+ t |= 1ULL << i;
+ }
+ }
+ const int col_blocks = THCCeilDiv(n_polys, threadsPerBlock);
+ dev_mask[cur_box_idx * col_blocks + col_start] = t;
+ }
+}
+
+// boxes is a N x 9 tensor
+at::Tensor poly_nms_cuda(const at::Tensor boxes, float nms_overlap_thresh) {
+
+ at::DeviceGuard guard(boxes.device());
+
+ using scalar_t = float;
+ AT_ASSERTM(boxes.device().is_cuda(), "boxes must be a CUDA tensor");
+ auto scores = boxes.select(1, 8);
+ auto order_t = std::get<1>(scores.sort(0, /*descending=*/true));
+ auto boxes_sorted = boxes.index_select(0, order_t);
+
+ int boxes_num = boxes.size(0);
+
+ const int col_blocks = THCCeilDiv(boxes_num, threadsPerBlock);
+
+ scalar_t* boxes_dev = boxes_sorted.data_ptr();
+
+ THCState *state = at::globalContext().lazyInitCUDA();
+
+ unsigned long long * mask_dev = NULL;
+
+ mask_dev = (unsigned long long *) THCudaMalloc(state, boxes_num * col_blocks * sizeof(unsigned long long));
+
+ dim3 blocks(THCCeilDiv(boxes_num, threadsPerBlock), THCCeilDiv(boxes_num, threadsPerBlock));
+ dim3 threads(threadsPerBlock);
+ poly_nms_kernel << > > (boxes_num, nms_overlap_thresh, boxes_dev, mask_dev);
+
+ std::vector mask_host(boxes_num * col_blocks);
+ THCudaCheck(cudaMemcpyAsync(
+ &mask_host[0],
+ mask_dev,
+ sizeof(unsigned long long) * boxes_num * col_blocks,
+ cudaMemcpyDeviceToHost,
+ at::cuda::getCurrentCUDAStream()
+ ));
+
+ std::vector remv(col_blocks);
+ memset(&remv[0], 0, sizeof(unsigned long long) * col_blocks);
+
+ at::Tensor keep = at::empty({ boxes_num }, boxes.options().dtype(at::kLong).device(at::kCPU));
+ int64_t * keep_out = keep.data_ptr();
+
+ int num_to_keep = 0;
+ for (int i = 0; i < boxes_num; i++) {
+ int nblock = i / threadsPerBlock;
+ int inblock = i % threadsPerBlock;
+
+ if (!(remv[nblock] & (1ULL << inblock))) {
+ keep_out[num_to_keep++] = i;
+ unsigned long long *p = &mask_host[0] + i * col_blocks;
+ for (int j = nblock; j < col_blocks; j++) {
+ remv[j] |= p[j];
+ }
+ }
+ }
+
+ THCudaFree(state, mask_dev);
+ return order_t.index({
+ keep.narrow(/*dim=*/0, /*start=*/0, /*length=*/num_to_keep).to(
+ order_t.device(), keep.scalar_type()) });
+}
+
diff --git a/utils/plots.py b/utils/plots.py
new file mode 100644
index 0000000..8120a54
--- /dev/null
+++ b/utils/plots.py
@@ -0,0 +1,540 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Plotting utils
+"""
+
+import math
+import os
+from copy import copy
+from pathlib import Path
+
+import cv2
+import matplotlib
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+import seaborn as sn
+import torch
+from PIL import Image, ImageDraw, ImageFont
+
+from utils.general import (LOGGER, Timeout, check_requirements, clip_coords, increment_path, is_ascii, is_chinese,
+ try_except, user_config_dir, xywh2xyxy, xyxy2xywh)
+from utils.metrics import fitness
+from utils.rboxs_utils import poly2hbb, poly2rbox, rbox2poly
+
+# Settings
+CONFIG_DIR = user_config_dir() # Ultralytics settings dir
+RANK = int(os.getenv('RANK', -1))
+matplotlib.rc('font', **{'size': 11})
+matplotlib.use('Agg') # for writing to files only
+
+
+class Colors:
+ # Ultralytics color palette https://ultralytics.com/
+ def __init__(self):
+ # hex = matplotlib.colors.TABLEAU_COLORS.values()
+ hex = ('FF3838', 'FF9D97', 'FF701F', 'FFB21D', 'CFD231', '48F90A', '92CC17', '3DDB86', '1A9334', '00D4BB',
+ '2C99A8', '00C2FF', '344593', '6473FF', '0018EC', '8438FF', '520085', 'CB38FF', 'FF95C8', 'FF37C7')
+ self.palette = [self.hex2rgb('#' + c) for c in hex]
+ self.n = len(self.palette)
+
+ def __call__(self, i, bgr=False):
+ c = self.palette[int(i) % self.n]
+ return (c[2], c[1], c[0]) if bgr else c
+
+ @staticmethod
+ def hex2rgb(h): # rgb order (PIL)
+ return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))
+
+
+colors = Colors() # create instance for 'from utils.plots import colors'
+
+
+def check_font(font='Arial.ttf', size=10):
+ # Return a PIL TrueType Font, downloading to CONFIG_DIR if necessary
+ font = Path(font)
+ font = font if font.exists() else (CONFIG_DIR / font.name)
+ try:
+ return ImageFont.truetype(str(font) if font.exists() else font.name, size)
+ except Exception as e: # download if missing
+ url = "https://ultralytics.com/assets/" + font.name
+ print(f'Downloading {url} to {font}...')
+ torch.hub.download_url_to_file(url, str(font), progress=False)
+ try:
+ return ImageFont.truetype(str(font), size)
+ except TypeError:
+ check_requirements('Pillow>=8.4.0') # known issue https://github.com/ultralytics/yolov5/issues/5374
+
+
+class Annotator:
+ if RANK in (-1, 0):
+ check_font() # download TTF if necessary
+
+ # YOLOv5 Annotator for train/val mosaics and jpgs and detect/hub inference annotations
+ def __init__(self, im, line_width=None, font_size=None, font='Arial.ttf', pil=False, example='abc'):
+ assert im.data.contiguous, 'Image not contiguous. Apply np.ascontiguousarray(im) to Annotator() input images.'
+ self.pil = pil or not is_ascii(example) or is_chinese(example)
+ if self.pil: # use PIL
+ self.im = im if isinstance(im, Image.Image) else Image.fromarray(im)
+ self.im_cv2 = im
+ self.draw = ImageDraw.Draw(self.im)
+ self.font = check_font(font='Arial.Unicode.ttf' if is_chinese(example) else font,
+ size=font_size or max(round(sum(self.im.size) / 2 * 0.035), 12))
+ else: # use cv2
+ self.im = im
+ self.im_cv2 = im
+ self.lw = line_width or max(round(sum(im.shape) / 2 * 0.003), 2) # line width
+
+ def box_label(self, box, label='', color=(128, 128, 128), txt_color=(255, 255, 255)):
+ # Add one xyxy box to image with label
+ if self.pil or not is_ascii(label):
+ self.draw.rectangle(box, width=self.lw, outline=color) # box
+ if label:
+ w, h = self.font.getsize(label) # text width, height
+ outside = box[1] - h >= 0 # label fits outside box
+ self.draw.rectangle([box[0],
+ box[1] - h if outside else box[1],
+ box[0] + w + 1,
+ box[1] + 1 if outside else box[1] + h + 1], fill=color)
+ # self.draw.text((box[0], box[1]), label, fill=txt_color, font=self.font, anchor='ls') # for PIL>8.0
+ self.draw.text((box[0], box[1] - h if outside else box[1]), label, fill=txt_color, font=self.font)
+ else: # cv2
+ p1, p2 = (int(box[0]), int(box[1])), (int(box[2]), int(box[3]))
+ cv2.rectangle(self.im, p1, p2, color, thickness=self.lw, lineType=cv2.LINE_AA)
+ if label:
+ tf = max(self.lw - 1, 1) # font thickness
+ w, h = cv2.getTextSize(label, 0, fontScale=self.lw / 3, thickness=tf)[0] # text width, height
+ outside = p1[1] - h - 3 >= 0 # label fits outside box
+ p2 = p1[0] + w, p1[1] - h - 3 if outside else p1[1] + h + 3
+ cv2.rectangle(self.im, p1, p2, color, -1, cv2.LINE_AA) # filled
+ cv2.putText(self.im, label, (p1[0], p1[1] - 2 if outside else p1[1] + h + 2), 0, self.lw / 3, txt_color,
+ thickness=tf, lineType=cv2.LINE_AA)
+
+ def poly_label(self, poly, label='', color=(128, 128, 128), txt_color=(255, 255, 255)):
+ # if self.pil or not is_ascii(label):
+ # self.draw.polygon(xy=poly, outline=color)
+ # if label:
+ # xmax, xmin, ymax, ymin = max(poly[0::2]), min(poly[0::2]), max(poly[1::2]), min(poly[1::2])
+ # x_label, y_label = (xmax + xmin)/2, (ymax + ymin)/2
+ # w, h = self.font.getsize(label) # text width, height
+ # outside = ymin - h >= 0 # label fits outside box
+ # self.draw.rectangle([x_label,
+ # y_label - h if outside else y_label,
+ # x_label + w + 1,
+ # y_label + 1 if outside else y_label + h + 1], fill=color)
+ # self.draw.text((x_label, y_label - h if outside else y_label), label, fill=txt_color, font=self.font)
+ # else:
+ if isinstance(poly, torch.Tensor):
+ poly = poly.cpu().numpy()
+ if isinstance(poly[0], torch.Tensor):
+ poly = [x.cpu().numpy() for x in poly]
+ polygon_list = np.array([(poly[0], poly[1]), (poly[2], poly[3]), \
+ (poly[4], poly[5]), (poly[6], poly[7])], np.int32)
+ cv2.drawContours(image=self.im_cv2, contours=[polygon_list], contourIdx=-1, color=color, thickness=self.lw)
+ if label:
+ tf = max(self.lw - 1, 1) # font thicknes
+ xmax, xmin, ymax, ymin = max(poly[0::2]), min(poly[0::2]), max(poly[1::2]), min(poly[1::2])
+ x_label, y_label = int((xmax + xmin)/2), int((ymax + ymin)/2)
+ w, h = cv2.getTextSize(label, 0, fontScale=self.lw / 3, thickness=tf)[0] # text width, height
+ cv2.rectangle(
+ self.im_cv2,
+ (x_label, y_label),
+ (x_label + w + 1, y_label + int(1.5*h)),
+ color, -1, cv2.LINE_AA
+ )
+ cv2.putText(self.im_cv2, label, (x_label, y_label + h), 0, self.lw / 3, txt_color, thickness=tf, lineType=cv2.LINE_AA)
+ self.im = self.im_cv2 if isinstance(self.im_cv2, Image.Image) else Image.fromarray(self.im_cv2)
+
+ def rectangle(self, xy, fill=None, outline=None, width=1):
+ # Add rectangle to image (PIL-only)
+ self.draw.rectangle(xy, fill, outline, width)
+
+ def text(self, xy, text, txt_color=(255, 255, 255)):
+ # Add text to image (PIL-only)
+ w, h = self.font.getsize(text) # text width, height
+ self.draw.text((xy[0], xy[1] - h + 1), text, fill=txt_color, font=self.font)
+
+ def result(self):
+ # Return annotated image as array
+ return np.asarray(self.im)
+
+
+def feature_visualization(x, module_type, stage, n=32, save_dir=Path('runs/detect/exp')):
+ """
+ x: Features to be visualized
+ module_type: Module type
+ stage: Module stage within model
+ n: Maximum number of feature maps to plot
+ save_dir: Directory to save results
+ """
+ if 'Detect' not in module_type:
+ batch, channels, height, width = x.shape # batch, channels, height, width
+ if height > 1 and width > 1:
+ f = save_dir / f"stage{stage}_{module_type.split('.')[-1]}_features.png" # filename
+
+ blocks = torch.chunk(x[0].cpu(), channels, dim=0) # select batch index 0, block by channels
+ n = min(n, channels) # number of plots
+ fig, ax = plt.subplots(math.ceil(n / 8), 8, tight_layout=True) # 8 rows x n/8 cols
+ ax = ax.ravel()
+ plt.subplots_adjust(wspace=0.05, hspace=0.05)
+ for i in range(n):
+ ax[i].imshow(blocks[i].squeeze()) # cmap='gray'
+ ax[i].axis('off')
+
+ print(f'Saving {f}... ({n}/{channels})')
+ plt.savefig(f, dpi=300, bbox_inches='tight')
+ plt.close()
+ np.save(str(f.with_suffix('.npy')), x[0].cpu().numpy()) # npy save
+
+
+def hist2d(x, y, n=100):
+ # 2d histogram used in labels.png and evolve.png
+ xedges, yedges = np.linspace(x.min(), x.max(), n), np.linspace(y.min(), y.max(), n)
+ hist, xedges, yedges = np.histogram2d(x, y, (xedges, yedges))
+ xidx = np.clip(np.digitize(x, xedges) - 1, 0, hist.shape[0] - 1)
+ yidx = np.clip(np.digitize(y, yedges) - 1, 0, hist.shape[1] - 1)
+ return np.log(hist[xidx, yidx])
+
+
+def butter_lowpass_filtfilt(data, cutoff=1500, fs=50000, order=5):
+ from scipy.signal import butter, filtfilt
+
+ # https://stackoverflow.com/questions/28536191/how-to-filter-smooth-with-scipy-numpy
+ def butter_lowpass(cutoff, fs, order):
+ nyq = 0.5 * fs
+ normal_cutoff = cutoff / nyq
+ return butter(order, normal_cutoff, btype='low', analog=False)
+
+ b, a = butter_lowpass(cutoff, fs, order=order)
+ return filtfilt(b, a, data) # forward-backward filter
+
+
+def output_to_target(output): #list*(n, [xylsθ, conf, cls]) θ ∈ [-pi/2, pi/2)
+ # Convert model output to target format [batch_id, class_id, x, y, l, s, theta, conf]
+ targets = []
+ for i, o in enumerate(output):
+ for *rbox, conf, cls in o.cpu().numpy():
+ targets.append([i, cls, *list(*(np.array(rbox)[None])), conf])
+ return np.array(targets)
+
+
+def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max_size=2048, max_subplots=4):
+ """
+ Args:
+ imgs (tensor): (b, 3, height, width)
+ targets_train (tensor): (n_targets, [batch_id clsid cx cy l s theta gaussian_θ_labels]) θ∈[-pi/2, pi/2)
+ targets_pred (array): (n, [batch_id, class_id, cx, cy, l, s, theta, conf]) θ∈[-pi/2, pi/2)
+ paths (list[str,...]): (b)
+ fname (str): (1)
+ names :
+
+ """
+ # Plot image grid with labels
+ if isinstance(images, torch.Tensor):
+ images = images.cpu().float().numpy()
+ if isinstance(targets, torch.Tensor):
+ targets = targets.cpu().numpy()
+ if np.max(images[0]) <= 1:
+ images *= 255 # de-normalise (optional)
+ bs, _, h, w = images.shape # batch size, _, height, width
+ bs = min(bs, max_subplots) # limit plot images
+ ns = np.ceil(bs ** 0.5) # number of subplots (square)
+
+ # Build Image
+ mosaic = np.full((int(ns * h), int(ns * w), 3), 255, dtype=np.uint8) # init
+ for i, im in enumerate(images):
+ if i == max_subplots: # if last batch has fewer images than we expect
+ break
+ x, y = int(w * (i // ns)), int(h * (i % ns)) # block origin
+ im = im.transpose(1, 2, 0)
+ mosaic[y:y + h, x:x + w, :] = im
+
+ # Resize (optional)
+ scale = max_size / ns / max(h, w)
+ if scale < 1:
+ h = math.ceil(scale * h)
+ w = math.ceil(scale * w)
+ mosaic = cv2.resize(mosaic, tuple(int(x * ns) for x in (w, h)))
+
+ # Annotate
+ fs = int((h + w) * ns * 0.01) # font size
+ annotator = Annotator(mosaic, line_width=round(fs / 10), font_size=fs, pil=True)
+ for i in range(i + 1):
+ x, y = int(w * (i // ns)), int(h * (i % ns)) # block origin
+ annotator.rectangle([x, y, x + w, y + h], None, (255, 255, 255), width=2) # borders
+ if paths:
+ annotator.text((x + 5, y + 5 + h), text=Path(paths[i]).name[:40], txt_color=(220, 220, 220)) # filenames
+ if len(targets) > 0:
+ ti = targets[targets[:, 0] == i] # image targets, (n, [img_index clsid cx cy l s theta gaussian_θ_labels])
+ # boxes = xywh2xyxy(ti[:, 2:6]).T
+ rboxes = ti[:, 2:7]
+ classes = ti[:, 1].astype('int')
+ # labels = ti.shape[1] == 6 # labels if no conf column
+ labels = ti.shape[1] == 187 # labels if no conf column
+ # conf = None if labels else ti[:, 6] # check for confidence presence (label vs pred)
+ conf = None if labels else ti[:, 7] # check for confidence presence (label vs pred)
+
+ # if boxes.shape[1]:
+ # if boxes.max() <= 1.01: # if normalized with tolerance 0.01
+ # boxes[[0, 2]] *= w # scale to pixels
+ # boxes[[1, 3]] *= h
+ # elif scale < 1: # absolute coords need scale if image scales
+ # boxes *= scale
+ polys = rbox2poly(rboxes)
+ if scale < 1:
+ polys *= scale
+ # boxes[[0, 2]] += x
+ # boxes[[1, 3]] += y
+ polys[:, [0, 2, 4, 6]] += x
+ polys[:, [1, 3, 5, 7]] += y
+ # for j, box in enumerate(boxes.T.tolist()):
+ # cls = classes[j]
+ # color = colors(cls)
+ # cls = names[cls] if names else cls
+ # if labels or conf[j] > 0.25: # 0.25 conf thresh
+ # label = f'{cls}' if labels else f'{cls} {conf[j]:.1f}'
+ # annotator.box_label(box, label, color=color)
+ for j, poly in enumerate(polys.tolist()):
+ cls = classes[j]
+ color = colors(cls)
+ cls = names[cls] if names else cls
+ if labels or conf[j] > 0.25: # 0.25 conf thresh
+ label = f'{cls}' if labels else f'{cls} {conf[j]:.1f}'
+ annotator.poly_label(poly, label, color=color)
+ annotator.im.save(fname) # save
+
+def plot_lr_scheduler(optimizer, scheduler, epochs=300, save_dir=''):
+ # Plot LR simulating training for full epochs
+ optimizer, scheduler = copy(optimizer), copy(scheduler) # do not modify originals
+ y = []
+ for _ in range(epochs):
+ scheduler.step()
+ y.append(optimizer.param_groups[0]['lr'])
+ plt.plot(y, '.-', label='LR')
+ plt.xlabel('epoch')
+ plt.ylabel('LR')
+ plt.grid()
+ plt.xlim(0, epochs)
+ plt.ylim(0)
+ plt.savefig(Path(save_dir) / 'LR.png', dpi=200)
+ plt.close()
+
+
+def plot_val_txt(): # from utils.plots import *; plot_val()
+ # Plot val.txt histograms
+ x = np.loadtxt('val.txt', dtype=np.float32)
+ box = xyxy2xywh(x[:, :4])
+ cx, cy = box[:, 0], box[:, 1]
+
+ fig, ax = plt.subplots(1, 1, figsize=(6, 6), tight_layout=True)
+ ax.hist2d(cx, cy, bins=600, cmax=10, cmin=0)
+ ax.set_aspect('equal')
+ plt.savefig('hist2d.png', dpi=300)
+
+ fig, ax = plt.subplots(1, 2, figsize=(12, 6), tight_layout=True)
+ ax[0].hist(cx, bins=600)
+ ax[1].hist(cy, bins=600)
+ plt.savefig('hist1d.png', dpi=200)
+
+
+def plot_targets_txt(): # from utils.plots import *; plot_targets_txt()
+ # Plot targets.txt histograms
+ x = np.loadtxt('targets.txt', dtype=np.float32).T
+ s = ['x targets', 'y targets', 'width targets', 'height targets']
+ fig, ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)
+ ax = ax.ravel()
+ for i in range(4):
+ ax[i].hist(x[i], bins=100, label=f'{x[i].mean():.3g} +/- {x[i].std():.3g}')
+ ax[i].legend()
+ ax[i].set_title(s[i])
+ plt.savefig('targets.jpg', dpi=200)
+
+
+def plot_val_study(file='', dir='', x=None): # from utils.plots import *; plot_val_study()
+ # Plot file=study.txt generated by val.py (or plot all study*.txt in dir)
+ save_dir = Path(file).parent if file else Path(dir)
+ plot2 = False # plot additional results
+ if plot2:
+ ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)[1].ravel()
+
+ fig2, ax2 = plt.subplots(1, 1, figsize=(8, 4), tight_layout=True)
+ # for f in [save_dir / f'study_coco_{x}.txt' for x in ['yolov5n6', 'yolov5s6', 'yolov5m6', 'yolov5l6', 'yolov5x6']]:
+ for f in sorted(save_dir.glob('study*.txt')):
+ y = np.loadtxt(f, dtype=np.float32, usecols=[0, 1, 2, 3, 7, 8, 9], ndmin=2).T
+ x = np.arange(y.shape[1]) if x is None else np.array(x)
+ if plot2:
+ s = ['P', 'R', 'HBBmAP@.5', 'HBBmAP@.5:.95', 't_preprocess (ms/img)', 't_inference (ms/img)', 't_NMS (ms/img)']
+ for i in range(7):
+ ax[i].plot(x, y[i], '.-', linewidth=2, markersize=8)
+ ax[i].set_title(s[i])
+
+ j = y[3].argmax() + 1
+ ax2.plot(y[5, 1:j], y[3, 1:j] * 1E2, '.-', linewidth=2, markersize=8,
+ label=f.stem.replace('study_coco_', '').replace('yolo', 'YOLO'))
+
+ ax2.plot(1E3 / np.array([209, 140, 97, 58, 35, 18]), [34.6, 40.5, 43.0, 47.5, 49.7, 51.5],
+ 'k.-', linewidth=2, markersize=8, alpha=.25, label='EfficientDet')
+
+ ax2.grid(alpha=0.2)
+ ax2.set_yticks(np.arange(20, 60, 5))
+ ax2.set_xlim(0, 57)
+ ax2.set_ylim(25, 55)
+ ax2.set_xlabel('GPU Speed (ms/img)')
+ ax2.set_ylabel('COCO AP val')
+ ax2.legend(loc='lower right')
+ f = save_dir / 'study.png'
+ print(f'Saving {f}...')
+ plt.savefig(f, dpi=300)
+
+
+@try_except # known issue https://github.com/ultralytics/yolov5/issues/5395
+@Timeout(30) # known issue https://github.com/ultralytics/yolov5/issues/5611
+def plot_labels(labels, names=(), save_dir=Path(''), img_size=1024):
+ rboxes = poly2rbox(labels[:, 1:])
+ labels = np.concatenate((labels[:, :1], rboxes[:, :-1]), axis=1) # [cls xyls]
+
+ # plot dataset labels
+ LOGGER.info(f"Plotting labels to {save_dir / 'labels_xyls.jpg'}... ")
+ c, b = labels[:, 0], labels[:, 1:].transpose() # classes, hboxes(xyls)
+ nc = int(c.max() + 1) # number of classes
+ x = pd.DataFrame(b.transpose(), columns=['x', 'y', 'long_edge', 'short_edge'])
+
+ # seaborn correlogram
+ sn.pairplot(x, corner=True, diag_kind='auto', kind='hist', diag_kws=dict(bins=50), plot_kws=dict(pmax=0.9))
+ plt.savefig(save_dir / 'labels_correlogram.jpg', dpi=200)
+ plt.close()
+
+ # matplotlib labels
+ matplotlib.use('svg') # faster
+ ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)[1].ravel()
+ y = ax[0].hist(c, bins=np.linspace(0, nc, nc + 1) - 0.5, rwidth=0.8)
+ # [y[2].patches[i].set_color([x / 255 for x in colors(i)]) for i in range(nc)] # update colors bug #3195
+ ax[0].set_ylabel('instances')
+ if 0 < len(names) < 30:
+ ax[0].set_xticks(range(len(names)))
+ ax[0].set_xticklabels(names, rotation=90, fontsize=10)
+ else:
+ ax[0].set_xlabel('classes')
+ sn.histplot(x, x='x', y='y', ax=ax[2], bins=50, pmax=0.9)
+ sn.histplot(x, x='long_edge', y='short_edge', ax=ax[3], bins=50, pmax=0.9)
+
+ # rectangles
+ # labels[:, 1:3] = 0.5 # center
+ labels[:, 1:3] = 0.5 * img_size # center
+ # labels[:, 1:] = xywh2xyxy(labels[:, 1:]) * 2000
+ labels[:, 1:] = xywh2xyxy(labels[:, 1:])
+ # img = Image.fromarray(np.ones((2000, 2000, 3), dtype=np.uint8) * 255)
+ img = Image.fromarray(np.ones((img_size, img_size, 3), dtype=np.uint8) * 255)
+ for cls, *box in labels[:1000]:
+ ImageDraw.Draw(img).rectangle(box, width=1, outline=colors(cls)) # plot
+ ax[1].imshow(img)
+ ax[1].axis('off')
+
+ for a in [0, 1, 2, 3]:
+ for s in ['top', 'right', 'left', 'bottom']:
+ ax[a].spines[s].set_visible(False)
+
+ plt.savefig(save_dir / 'labels_xyls.jpg', dpi=200)
+ matplotlib.use('Agg')
+ plt.close()
+
+
+def plot_evolve(evolve_csv='path/to/evolve.csv'): # from utils.plots import *; plot_evolve()
+ # Plot evolve.csv hyp evolution results
+ evolve_csv = Path(evolve_csv)
+ data = pd.read_csv(evolve_csv)
+ keys = [x.strip() for x in data.columns]
+ x = data.values
+ f = fitness(x)
+ j = np.argmax(f) # max fitness index
+ plt.figure(figsize=(10, 12), tight_layout=True)
+ matplotlib.rc('font', **{'size': 8})
+ for i, k in enumerate(keys[7:]):
+ v = x[:, 7 + i]
+ mu = v[j] # best single result
+ plt.subplot(6, 5, i + 1)
+ plt.scatter(v, f, c=hist2d(v, f, 20), cmap='viridis', alpha=.8, edgecolors='none')
+ plt.plot(mu, f.max(), 'k+', markersize=15)
+ plt.title(f'{k} = {mu:.3g}', fontdict={'size': 9}) # limit to 40 characters
+ if i % 5 != 0:
+ plt.yticks([])
+ print(f'{k:>15}: {mu:.3g}')
+ f = evolve_csv.with_suffix('.png') # filename
+ plt.savefig(f, dpi=200)
+ plt.close()
+ print(f'Saved {f}')
+
+
+def plot_results(file='path/to/results.csv', dir=''):
+ # Plot training results.csv. Usage: from utils.plots import *; plot_results('path/to/results.csv')
+ save_dir = Path(file).parent if file else Path(dir)
+ #fig, ax = plt.subplots(2, 5, figsize=(12, 6), tight_layout=True)
+ fig, ax = plt.subplots(2, 6, figsize=(18, 6), tight_layout=True)
+ ax = ax.ravel()
+ files = list(save_dir.glob('results*.csv'))
+ assert len(files), f'No results.csv files found in {save_dir.resolve()}, nothing to plot.'
+ for fi, f in enumerate(files):
+ try:
+ data = pd.read_csv(f)
+ s = [x.strip() for x in data.columns]
+ x = data.values[:, 0]
+ #for i, j in enumerate([1, 2, 3, 4, 5, 8, 9, 10, 6, 7]):
+ for i, j in enumerate([1, 2, 3, 4, 5, 6, 9, 10, 11, 12, 7, 8]):
+ y = data.values[:, j]
+ # y[y == 0] = np.nan # don't show zero values
+ ax[i].plot(x, y, marker='.', label=f.stem, linewidth=2, markersize=8)
+ ax[i].set_title(s[j], fontsize=12)
+ # if j in [8, 9, 10]: # share train and val loss y axes
+ # ax[i].get_shared_y_axes().join(ax[i], ax[i - 5])
+ except Exception as e:
+ print(f'Warning: Plotting error for {f}: {e}')
+ ax[1].legend()
+ fig.savefig(save_dir / 'results.png', dpi=200)
+ plt.close()
+
+
+def profile_idetection(start=0, stop=0, labels=(), save_dir=''):
+ # Plot iDetection '*.txt' per-image logs. from utils.plots import *; profile_idetection()
+ ax = plt.subplots(2, 4, figsize=(12, 6), tight_layout=True)[1].ravel()
+ s = ['Images', 'Free Storage (GB)', 'RAM Usage (GB)', 'Battery', 'dt_raw (ms)', 'dt_smooth (ms)', 'real-world FPS']
+ files = list(Path(save_dir).glob('frames*.txt'))
+ for fi, f in enumerate(files):
+ try:
+ results = np.loadtxt(f, ndmin=2).T[:, 90:-30] # clip first and last rows
+ n = results.shape[1] # number of rows
+ x = np.arange(start, min(stop, n) if stop else n)
+ results = results[:, x]
+ t = (results[0] - results[0].min()) # set t0=0s
+ results[0] = x
+ for i, a in enumerate(ax):
+ if i < len(results):
+ label = labels[fi] if len(labels) else f.stem.replace('frames_', '')
+ a.plot(t, results[i], marker='.', label=label, linewidth=1, markersize=5)
+ a.set_title(s[i])
+ a.set_xlabel('time (s)')
+ # if fi == len(files) - 1:
+ # a.set_ylim(bottom=0)
+ for side in ['top', 'right']:
+ a.spines[side].set_visible(False)
+ else:
+ a.remove()
+ except Exception as e:
+ print(f'Warning: Plotting error for {f}; {e}')
+ ax[1].legend()
+ plt.savefig(Path(save_dir) / 'idetection_profile.png', dpi=200)
+
+
+def save_one_box(xyxy, im, file='image.jpg', gain=1.02, pad=10, square=False, BGR=False, save=True):
+ # Save image crop as {file} with crop size multiple {gain} and {pad} pixels. Save and/or return crop
+ xyxy = torch.tensor(xyxy).view(-1, 4)
+ b = xyxy2xywh(xyxy) # boxes
+ if square:
+ b[:, 2:] = b[:, 2:].max(1)[0].unsqueeze(1) # attempt rectangle to square
+ b[:, 2:] = b[:, 2:] * gain + pad # box wh * gain + pad
+ xyxy = xywh2xyxy(b).long()
+ clip_coords(xyxy, im.shape)
+ crop = im[int(xyxy[0, 1]):int(xyxy[0, 3]), int(xyxy[0, 0]):int(xyxy[0, 2]), ::(1 if BGR else -1)]
+ if save:
+ file.parent.mkdir(parents=True, exist_ok=True) # make directory
+ cv2.imwrite(str(increment_path(file).with_suffix('.jpg')), crop)
+ return crop
diff --git a/utils/rboxs_utils.py b/utils/rboxs_utils.py
new file mode 100644
index 0000000..1e50a2d
--- /dev/null
+++ b/utils/rboxs_utils.py
@@ -0,0 +1,200 @@
+"""
+Oriented Bounding Boxes utils
+"""
+import numpy as np
+pi = 3.141592
+import cv2
+import torch
+
+def gaussian_label_cpu(label, num_class, u=0, sig=4.0):
+ """
+ 转换成CSL Labels:
+ 用高斯窗口函数根据角度θ的周期性赋予gt labels同样的周期性,使得损失函数在计算边界处时可以做到“差值很大但loss很小”;
+ 并且使得其labels具有环形特征,能够反映各个θ之间的角度距离
+ Args:
+ label (float32):[1], theta class
+ num_theta_class (int): [1], theta class num
+ u (float32):[1], μ in gaussian function
+ sig (float32):[1], σ in gaussian function, which is window radius for Circular Smooth Label
+ Returns:
+ csl_label (array): [num_theta_class], gaussian function smooth label
+ """
+ x = np.arange(-num_class/2, num_class/2)
+ y_sig = np.exp(-(x - u) ** 2 / (2 * sig ** 2))
+ index = int(num_class/2 - label)
+ return np.concatenate([y_sig[index:],
+ y_sig[:index]], axis=0)
+
+def regular_theta(theta, mode='180', start=-pi/2):
+ """
+ limit theta ∈ [-pi/2, pi/2)
+ """
+ assert mode in ['360', '180']
+ cycle = 2 * pi if mode == '360' else pi
+
+ theta = theta - start
+ theta = theta % cycle
+ return theta + start
+
+def poly2rbox(polys, num_cls_thata=180, radius=6.0, use_pi=False, use_gaussian=False):
+ """
+ Trans poly format to rbox format.
+ Args:
+ polys (array): (num_gts, [x1 y1 x2 y2 x3 y3 x4 y4])
+ num_cls_thata (int): [1], theta class num
+ radius (float32): [1], window radius for Circular Smooth Label
+ use_pi (bool): True θ∈[-pi/2, pi/2) , False θ∈[0, 180)
+
+ Returns:
+ use_gaussian True:
+ rboxes (array):
+ csl_labels (array): (num_gts, num_cls_thata)
+ elif
+ rboxes (array): (num_gts, [cx cy l s θ])
+ """
+ assert polys.shape[-1] == 8
+ if use_gaussian:
+ csl_labels = []
+ rboxes = []
+ for poly in polys:
+ poly = np.float32(poly.reshape(4, 2))
+ (x, y), (w, h), angle = cv2.minAreaRect(poly) # θ ∈ [0, 90]
+ angle = -angle # θ ∈ [-90, 0]
+ theta = angle / 180 * pi # 转为pi制
+
+ # trans opencv format to longedge format θ ∈ [-pi/2, pi/2]
+ if w != max(w, h):
+ w, h = h, w
+ theta += pi/2
+ theta = regular_theta(theta) # limit theta ∈ [-pi/2, pi/2)
+ angle = (theta * 180 / pi) + 90 # θ ∈ [0, 180)
+
+ if not use_pi: # 采用angle弧度制 θ ∈ [0, 180)
+ rboxes.append([x, y, w, h, angle])
+ else: # 采用pi制
+ rboxes.append([x, y, w, h, theta])
+ if use_gaussian:
+ csl_label = gaussian_label_cpu(label=angle, num_class=num_cls_thata, u=0, sig=radius)
+ csl_labels.append(csl_label)
+ if use_gaussian:
+ return np.array(rboxes), np.array(csl_labels)
+ return np.array(rboxes)
+
+# def rbox2poly(rboxes):
+# """
+# Trans rbox format to poly format.
+# Args:
+# rboxes (array): (num_gts, [cx cy l s θ]) θ∈(0, 180]
+
+# Returns:
+# polys (array): (num_gts, [x1 y1 x2 y2 x3 y3 x4 y4])
+# """
+# assert rboxes.shape[-1] == 5
+# polys = []
+# for rbox in rboxes:
+# x, y, w, h, theta = rbox
+# if theta > 90 and theta <= 180: # longedge format -> opencv format
+# w, h = h, w
+# theta -= 90
+# if theta <= 0 or theta > 90:
+# print("cv2.minAreaRect occurs some error. θ isn't in range(0, 90]. The longedge format is: ", rbox)
+
+# poly = cv2.boxPoints(((x, y), (w, h), theta)).reshape(-1)
+# polys.append(poly)
+# return np.array(polys)
+
+def rbox2poly(obboxes):
+ """
+ Trans rbox format to poly format.
+ Args:
+ rboxes (array/tensor): (num_gts, [cx cy l s θ]) θ∈[-pi/2, pi/2)
+
+ Returns:
+ polys (array/tensor): (num_gts, [x1 y1 x2 y2 x3 y3 x4 y4])
+ """
+ if isinstance(obboxes, torch.Tensor):
+ center, w, h, theta = obboxes[:, :2], obboxes[:, 2:3], obboxes[:, 3:4], obboxes[:, 4:5]
+ Cos, Sin = torch.cos(theta), torch.sin(theta)
+
+ vector1 = torch.cat(
+ (w/2 * Cos, -w/2 * Sin), dim=-1)
+ vector2 = torch.cat(
+ (-h/2 * Sin, -h/2 * Cos), dim=-1)
+ point1 = center + vector1 + vector2
+ point2 = center + vector1 - vector2
+ point3 = center - vector1 - vector2
+ point4 = center - vector1 + vector2
+ order = obboxes.shape[:-1]
+ return torch.cat(
+ (point1, point2, point3, point4), dim=-1).reshape(*order, 8)
+ else:
+ center, w, h, theta = np.split(obboxes, (2, 3, 4), axis=-1)
+ Cos, Sin = np.cos(theta), np.sin(theta)
+
+ vector1 = np.concatenate(
+ [w/2 * Cos, -w/2 * Sin], axis=-1)
+ vector2 = np.concatenate(
+ [-h/2 * Sin, -h/2 * Cos], axis=-1)
+
+ point1 = center + vector1 + vector2
+ point2 = center + vector1 - vector2
+ point3 = center - vector1 - vector2
+ point4 = center - vector1 + vector2
+ order = obboxes.shape[:-1]
+ return np.concatenate(
+ [point1, point2, point3, point4], axis=-1).reshape(*order, 8)
+
+def poly2hbb(polys):
+ """
+ Trans poly format to hbb format
+ Args:
+ rboxes (array/tensor): (num_gts, poly)
+
+ Returns:
+ hbboxes (array/tensor): (num_gts, [xc yc w h])
+ """
+ assert polys.shape[-1] == 8
+ if isinstance(polys, torch.Tensor):
+ x = polys[:, 0::2] # (num, 4)
+ y = polys[:, 1::2]
+ x_max = torch.amax(x, dim=1) # (num)
+ x_min = torch.amin(x, dim=1)
+ y_max = torch.amax(y, dim=1)
+ y_min = torch.amin(y, dim=1)
+ x_ctr, y_ctr = (x_max + x_min) / 2.0, (y_max + y_min) / 2.0 # (num)
+ h = y_max - y_min # (num)
+ w = x_max - x_min
+ x_ctr, y_ctr, w, h = x_ctr.reshape(-1, 1), y_ctr.reshape(-1, 1), w.reshape(-1, 1), h.reshape(-1, 1) # (num, 1)
+ hbboxes = torch.cat((x_ctr, y_ctr, w, h), dim=1)
+ else:
+ x = polys[:, 0::2] # (num, 4)
+ y = polys[:, 1::2]
+ x_max = np.amax(x, axis=1) # (num)
+ x_min = np.amin(x, axis=1)
+ y_max = np.amax(y, axis=1)
+ y_min = np.amin(y, axis=1)
+ x_ctr, y_ctr = (x_max + x_min) / 2.0, (y_max + y_min) / 2.0 # (num)
+ h = y_max - y_min # (num)
+ w = x_max - x_min
+ x_ctr, y_ctr, w, h = x_ctr.reshape(-1, 1), y_ctr.reshape(-1, 1), w.reshape(-1, 1), h.reshape(-1, 1) # (num, 1)
+ hbboxes = np.concatenate((x_ctr, y_ctr, w, h), axis=1)
+ return hbboxes
+
+def poly_filter(polys, h, w):
+ """
+ Filter the poly labels which is out of the image.
+ Args:
+ polys (array): (num, 8)
+
+ Return:
+ keep_masks (array): (num)
+ """
+ x = polys[:, 0::2] # (num, 4)
+ y = polys[:, 1::2]
+ x_max = np.amax(x, axis=1) # (num)
+ x_min = np.amin(x, axis=1)
+ y_max = np.amax(y, axis=1)
+ y_min = np.amin(y, axis=1)
+ x_ctr, y_ctr = (x_max + x_min) / 2.0, (y_max + y_min) / 2.0 # (num)
+ keep_masks = (x_ctr > 0) & (x_ctr < w) & (y_ctr > 0) & (y_ctr < h)
+ return keep_masks
\ No newline at end of file
diff --git a/utils/torch_utils.py b/utils/torch_utils.py
new file mode 100644
index 0000000..cddb173
--- /dev/null
+++ b/utils/torch_utils.py
@@ -0,0 +1,318 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+PyTorch utils
+"""
+
+import datetime
+import math
+import os
+import platform
+import subprocess
+import time
+from contextlib import contextmanager
+from copy import deepcopy
+from pathlib import Path
+
+import torch
+import torch.distributed as dist
+import torch.nn as nn
+import torch.nn.functional as F
+
+from utils.general import LOGGER
+
+try:
+ import thop # for FLOPs computation
+except ImportError:
+ thop = None
+
+
+@contextmanager
+def torch_distributed_zero_first(local_rank: int):
+ """
+ Decorator to make all processes in distributed training wait for each local_master to do something.
+ """
+ if local_rank not in [-1, 0]:
+ dist.barrier(device_ids=[local_rank])
+ yield
+ if local_rank == 0:
+ dist.barrier(device_ids=[0])
+
+
+def date_modified(path=__file__):
+ # return human-readable file modification date, i.e. '2021-3-26'
+ t = datetime.datetime.fromtimestamp(Path(path).stat().st_mtime)
+ return f'{t.year}-{t.month}-{t.day}'
+
+
+def git_describe(path=Path(__file__).parent): # path must be a directory
+ # return human-readable git description, i.e. v5.0-5-g3e25f1e https://git-scm.com/docs/git-describe
+ s = f'git -C {path} describe --tags --long --always'
+ try:
+ return subprocess.check_output(s, shell=True, stderr=subprocess.STDOUT).decode()[:-1]
+ except subprocess.CalledProcessError as e:
+ return '' # not a git repository
+
+
+def select_device(device='', batch_size=0, newline=True):
+ # device = 'cpu' or '0' or '0,1,2,3'
+ s = f'YOLOv5 🚀 {git_describe() or date_modified()} torch {torch.__version__} ' # string
+ device = str(device).strip().lower().replace('cuda:', '') # to string, 'cuda:0' to '0'
+ cpu = device == 'cpu'
+ if cpu:
+ os.environ['CUDA_VISIBLE_DEVICES'] = '-1' # force torch.cuda.is_available() = False
+ elif device: # non-cpu device requested
+ os.environ['CUDA_VISIBLE_DEVICES'] = device # set environment variable
+ assert torch.cuda.is_available(), f'CUDA unavailable, invalid device {device} requested' # check availability
+
+ cuda = not cpu and torch.cuda.is_available()
+ if cuda:
+ devices = device.split(',') if device else '0' # range(torch.cuda.device_count()) # i.e. 0,1,6,7
+ n = len(devices) # device count
+ if n > 1 and batch_size > 0: # check batch_size is divisible by device_count
+ assert batch_size % n == 0, f'batch-size {batch_size} not multiple of GPU count {n}'
+ space = ' ' * (len(s) + 1)
+ for i, d in enumerate(devices):
+ p = torch.cuda.get_device_properties(i)
+ s += f"{'' if i == 0 else space}CUDA:{d} ({p.name}, {p.total_memory / 1024 ** 2:.0f}MiB)\n" # bytes to MB
+ else:
+ s += 'CPU\n'
+
+ if not newline:
+ s = s.rstrip()
+ LOGGER.info(s.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else s) # emoji-safe
+ return torch.device('cuda:0' if cuda else 'cpu')
+
+
+def time_sync():
+ # pytorch-accurate time
+ if torch.cuda.is_available():
+ torch.cuda.synchronize()
+ return time.time()
+
+
+def profile(input, ops, n=10, device=None):
+ # YOLOv5 speed/memory/FLOPs profiler
+ #
+ # Usage:
+ # input = torch.randn(16, 3, 640, 640)
+ # m1 = lambda x: x * torch.sigmoid(x)
+ # m2 = nn.SiLU()
+ # profile(input, [m1, m2], n=100) # profile over 100 iterations
+
+ results = []
+ device = device or select_device()
+ print(f"{'Params':>12s}{'GFLOPs':>12s}{'GPU_mem (GB)':>14s}{'forward (ms)':>14s}{'backward (ms)':>14s}"
+ f"{'input':>24s}{'output':>24s}")
+
+ for x in input if isinstance(input, list) else [input]:
+ x = x.to(device)
+ x.requires_grad = True
+ for m in ops if isinstance(ops, list) else [ops]:
+ m = m.to(device) if hasattr(m, 'to') else m # device
+ m = m.half() if hasattr(m, 'half') and isinstance(x, torch.Tensor) and x.dtype is torch.float16 else m
+ tf, tb, t = 0, 0, [0, 0, 0] # dt forward, backward
+ try:
+ flops = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2 # GFLOPs
+ except:
+ flops = 0
+
+ try:
+ for _ in range(n):
+ t[0] = time_sync()
+ y = m(x)
+ t[1] = time_sync()
+ try:
+ _ = (sum(yi.sum() for yi in y) if isinstance(y, list) else y).sum().backward()
+ t[2] = time_sync()
+ except Exception as e: # no backward method
+ # print(e) # for debug
+ t[2] = float('nan')
+ tf += (t[1] - t[0]) * 1000 / n # ms per op forward
+ tb += (t[2] - t[1]) * 1000 / n # ms per op backward
+ mem = torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0 # (GB)
+ s_in = tuple(x.shape) if isinstance(x, torch.Tensor) else 'list'
+ s_out = tuple(y.shape) if isinstance(y, torch.Tensor) else 'list'
+ p = sum(list(x.numel() for x in m.parameters())) if isinstance(m, nn.Module) else 0 # parameters
+ print(f'{p:12}{flops:12.4g}{mem:>14.3f}{tf:14.4g}{tb:14.4g}{str(s_in):>24s}{str(s_out):>24s}')
+ results.append([p, flops, mem, tf, tb, s_in, s_out])
+ except Exception as e:
+ print(e)
+ results.append(None)
+ torch.cuda.empty_cache()
+ return results
+
+
+def is_parallel(model):
+ # Returns True if model is of type DP or DDP
+ return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel)
+
+
+def de_parallel(model):
+ # De-parallelize a model: returns single-GPU model if model is of type DP or DDP
+ return model.module if is_parallel(model) else model
+
+
+def initialize_weights(model):
+ for m in model.modules():
+ t = type(m)
+ if t is nn.Conv2d:
+ pass # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+ elif t is nn.BatchNorm2d:
+ m.eps = 1e-3
+ m.momentum = 0.03
+ elif t in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]:
+ m.inplace = True
+
+
+def find_modules(model, mclass=nn.Conv2d):
+ # Finds layer indices matching module class 'mclass'
+ return [i for i, m in enumerate(model.module_list) if isinstance(m, mclass)]
+
+
+def sparsity(model):
+ # Return global model sparsity
+ a, b = 0, 0
+ for p in model.parameters():
+ a += p.numel()
+ b += (p == 0).sum()
+ return b / a
+
+
+def prune(model, amount=0.3):
+ # Prune model to requested global sparsity
+ import torch.nn.utils.prune as prune
+ print('Pruning model... ', end='')
+ for name, m in model.named_modules():
+ if isinstance(m, nn.Conv2d):
+ prune.l1_unstructured(m, name='weight', amount=amount) # prune
+ prune.remove(m, 'weight') # make permanent
+ print(' %.3g global sparsity' % sparsity(model))
+
+
+def fuse_conv_and_bn(conv, bn):
+ # Fuse convolution and batchnorm layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/
+ fusedconv = nn.Conv2d(conv.in_channels,
+ conv.out_channels,
+ kernel_size=conv.kernel_size,
+ stride=conv.stride,
+ padding=conv.padding,
+ groups=conv.groups,
+ bias=True).requires_grad_(False).to(conv.weight.device)
+
+ # prepare filters
+ w_conv = conv.weight.clone().view(conv.out_channels, -1)
+ w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var)))
+ fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.shape))
+
+ # prepare spatial bias
+ b_conv = torch.zeros(conv.weight.size(0), device=conv.weight.device) if conv.bias is None else conv.bias
+ b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
+ fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)
+
+ return fusedconv
+
+
+def model_info(model, verbose=False, img_size=640):
+ # Model information. img_size may be int or list, i.e. img_size=640 or img_size=[640, 320]
+ n_p = sum(x.numel() for x in model.parameters()) # number parameters
+ n_g = sum(x.numel() for x in model.parameters() if x.requires_grad) # number gradients
+ if verbose:
+ print(f"{'layer':>5} {'name':>40} {'gradient':>9} {'parameters':>12} {'shape':>20} {'mu':>10} {'sigma':>10}")
+ for i, (name, p) in enumerate(model.named_parameters()):
+ name = name.replace('module_list.', '')
+ print('%5g %40s %9s %12g %20s %10.3g %10.3g' %
+ (i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))
+
+ try: # FLOPs
+ from thop import profile
+ stride = max(int(model.stride.max()), 32) if hasattr(model, 'stride') else 32
+ img = torch.zeros((1, model.yaml.get('ch', 3), stride, stride), device=next(model.parameters()).device) # input
+ flops = profile(deepcopy(model), inputs=(img,), verbose=False)[0] / 1E9 * 2 # stride GFLOPs
+ img_size = img_size if isinstance(img_size, list) else [img_size, img_size] # expand if int/float
+ fs = ', %.1f GFLOPs' % (flops * img_size[0] / stride * img_size[1] / stride) # 640x640 GFLOPs
+ except (ImportError, Exception):
+ fs = ''
+
+ LOGGER.info(f"Model Summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}")
+
+
+def scale_img(img, ratio=1.0, same_shape=False, gs=32): # img(16,3,256,416)
+ # scales img(bs,3,y,x) by ratio constrained to gs-multiple
+ if ratio == 1.0:
+ return img
+ else:
+ h, w = img.shape[2:]
+ s = (int(h * ratio), int(w * ratio)) # new size
+ img = F.interpolate(img, size=s, mode='bilinear', align_corners=False) # resize
+ if not same_shape: # pad/crop img
+ h, w = (math.ceil(x * ratio / gs) * gs for x in (h, w))
+ return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447) # value = imagenet mean
+
+
+def copy_attr(a, b, include=(), exclude=()):
+ # Copy attributes from b to a, options to only include [...] and to exclude [...]
+ for k, v in b.__dict__.items():
+ if (len(include) and k not in include) or k.startswith('_') or k in exclude:
+ continue
+ else:
+ setattr(a, k, v)
+
+
+class EarlyStopping:
+ # YOLOv5 simple early stopper
+ def __init__(self, patience=30):
+ self.best_fitness = 0.0 # i.e. mAP
+ self.best_epoch = 0
+ self.patience = patience or float('inf') # epochs to wait after fitness stops improving to stop
+ self.possible_stop = False # possible stop may occur next epoch
+
+ def __call__(self, epoch, fitness):
+ if fitness >= self.best_fitness: # >= 0 to allow for early zero-fitness stage of training
+ self.best_epoch = epoch
+ self.best_fitness = fitness
+ delta = epoch - self.best_epoch # epochs without improvement
+ self.possible_stop = delta >= (self.patience - 1) # possible stop may occur next epoch
+ stop = delta >= self.patience # stop training if patience exceeded
+ if stop:
+ LOGGER.info(f'Stopping training early as no improvement observed in last {self.patience} epochs. '
+ f'Best results observed at epoch {self.best_epoch}, best model saved as best.pt.\n'
+ f'To update EarlyStopping(patience={self.patience}) pass a new patience value, '
+ f'i.e. `python train.py --patience 300` or use `--patience 0` to disable EarlyStopping.')
+ return stop
+
+
+class ModelEMA:
+ """ Model Exponential Moving Average from https://github.com/rwightman/pytorch-image-models
+ Keep a moving average of everything in the model state_dict (parameters and buffers).
+ This is intended to allow functionality like
+ https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
+ A smoothed version of the weights is necessary for some training schemes to perform well.
+ This class is sensitive where it is initialized in the sequence of model init,
+ GPU assignment and distributed training wrappers.
+ """
+
+ def __init__(self, model, decay=0.9999, updates=0):
+ # Create EMA
+ self.ema = deepcopy(model.module if is_parallel(model) else model).eval() # FP32 EMA
+ # if next(model.parameters()).device.type != 'cpu':
+ # self.ema.half() # FP16 EMA
+ self.updates = updates # number of EMA updates
+ self.decay = lambda x: decay * (1 - math.exp(-x / 2000)) # decay exponential ramp (to help early epochs)
+ for p in self.ema.parameters():
+ p.requires_grad_(False)
+
+ def update(self, model):
+ # Update EMA parameters
+ with torch.no_grad():
+ self.updates += 1
+ d = self.decay(self.updates)
+
+ msd = model.module.state_dict() if is_parallel(model) else model.state_dict() # model state_dict
+ for k, v in self.ema.state_dict().items():
+ if v.dtype.is_floating_point:
+ v *= d
+ v += (1 - d) * msd[k].detach()
+
+ def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
+ # Update EMA attributes
+ copy_attr(self.ema, model, include, exclude)
diff --git a/val.py b/val.py
new file mode 100644
index 0000000..4804b75
--- /dev/null
+++ b/val.py
@@ -0,0 +1,400 @@
+# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
+"""
+Validate a trained YOLOv5 model accuracy on a custom dataset
+
+Usage:
+ $ python path/to/val.py --data coco128.yaml --weights yolov5s.pt --img 640
+"""
+
+import argparse
+import json
+import os
+import sys
+from pathlib import Path
+from threading import Thread
+
+import numpy as np
+import torch
+from tqdm import tqdm
+
+from utils.rboxs_utils import poly2hbb, rbox2poly
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[0] # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+ sys.path.append(str(ROOT)) # add ROOT to PATH
+ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
+
+from models.common import DetectMultiBackend
+from utils.callbacks import Callbacks
+from utils.datasets import create_dataloader
+from utils.general import (LOGGER, box_iou, check_dataset, check_img_size, check_requirements, check_yaml,
+ coco80_to_coco91_class, colorstr, increment_path, non_max_suppression, print_args,
+ scale_coords, scale_polys, xywh2xyxy, xyxy2xywh, non_max_suppression_obb)
+from utils.metrics import ConfusionMatrix, ap_per_class
+from utils.plots import output_to_target, plot_images, plot_val_study
+from utils.torch_utils import select_device, time_sync
+
+
+def save_one_txt(predn, save_conf, shape, file):
+ # Save one txt result
+ gn = torch.tensor(shape)[[1, 0, 1, 0]] # normalization gain whwh
+ for *xyxy, conf, cls in predn.tolist():
+ xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
+ line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
+ with open(file, 'a') as f:
+ f.write(('%g ' * len(line)).rstrip() % line + '\n')
+
+
+# def save_one_json(predn, jdict, path, class_map):
+def save_one_json(pred_hbbn, pred_polyn, jdict, path, class_map):
+ """
+ Save one JSON result {"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236, "poly": [...]}
+ Args:
+ pred_hbbn (tensor): (n, [poly, conf, cls])
+ pred_polyn (tensor): (n, [xyxy, conf, cls])
+ """
+ image_id = int(path.stem) if path.stem.isnumeric() else path.stem
+ box = xyxy2xywh(pred_hbbn[:, :4]) # xywh
+ box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner
+ for p, b in zip(pred_polyn.tolist(), box.tolist()):
+ jdict.append({'image_id': image_id,
+ 'category_id': class_map[int(p[-1]) + 1], # COCO's category_id start from 1, not 0
+ 'bbox': [round(x, 1) for x in b],
+ 'score': round(p[-2], 5),
+ 'poly': [round(x, 1) for x in p[:8]],
+ 'file_name': path.stem})
+
+
+def process_batch(detections, labels, iouv):
+ """
+ Return correct predictions matrix. Both sets of boxes are in (x1, y1, x2, y2) format.
+ Arguments:
+ detections (Array[N, 6]), x1, y1, x2, y2, conf, class
+ labels (Array[M, 5]), class, x1, y1, x2, y2
+ Returns:
+ correct (Array[N, 10]), for 10 IoU levels
+ """
+ correct = torch.zeros(detections.shape[0], iouv.shape[0], dtype=torch.bool, device=iouv.device)
+ iou = box_iou(labels[:, 1:], detections[:, :4])
+ x = torch.where((iou >= iouv[0]) & (labels[:, 0:1] == detections[:, 5])) # IoU above threshold and classes match
+ if x[0].shape[0]:
+ matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy() # [label, detection, iou]
+ if x[0].shape[0] > 1:
+ matches = matches[matches[:, 2].argsort()[::-1]]
+ matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
+ # matches = matches[matches[:, 2].argsort()[::-1]]
+ matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
+ matches = torch.Tensor(matches).to(iouv.device)
+ correct[matches[:, 1].long()] = matches[:, 2:3] >= iouv
+ return correct
+
+
+@torch.no_grad()
+def run(data,
+ weights=None, # model.pt path(s)
+ batch_size=32, # batch size
+ imgsz=640, # inference size (pixels)
+ conf_thres=0.01, # confidence threshold
+ iou_thres=0.4, # NMS IoU threshold
+ task='val', # train, val, test, speed or study
+ device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
+ workers=8, # max dataloader workers (per RANK in DDP mode)
+ single_cls=False, # treat as single-class dataset
+ augment=False, # augmented inference
+ verbose=False, # verbose output
+ save_txt=False, # save results to *.txt
+ save_hybrid=False, # save label+prediction hybrid results to *.txt
+ save_conf=False, # save confidences in --save-txt labels
+ save_json=False, # save a COCO-JSON results file
+ project=ROOT / 'runs/val', # save to project/name
+ name='exp', # save to project/name
+ exist_ok=False, # existing project/name ok, do not increment
+ half=True, # use FP16 half-precision inference
+ dnn=False, # use OpenCV DNN for ONNX inference
+ model=None,
+ dataloader=None,
+ save_dir=Path(''),
+ plots=True,
+ callbacks=Callbacks(),
+ compute_loss=None,
+ ):
+ # Initialize/load model and set device
+ training = model is not None
+ if training: # called by train.py
+ device, pt, jit, engine = next(model.parameters()).device, True, False, False # get model device, PyTorch model
+
+ half &= device.type != 'cpu' # half precision only supported on CUDA
+ model.half() if half else model.float()
+ else: # called directly
+ device = select_device(device, batch_size=batch_size)
+
+ # Directories
+ save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
+ (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
+
+ # Load model
+ model = DetectMultiBackend(weights, device=device, dnn=dnn)
+ stride, pt, jit, engine = model.stride, model.pt, model.jit, model.engine
+ imgsz = check_img_size(imgsz, s=stride) # check image size
+ half &= (pt or jit or engine) and device.type != 'cpu' # half precision only supported by PyTorch on CUDA
+ if pt or jit:
+ model.model.half() if half else model.model.float()
+ elif engine:
+ batch_size = model.batch_size
+ else:
+ half = False
+ batch_size = 1 # export.py models default to batch-size 1
+ device = torch.device('cpu')
+ LOGGER.info(f'Forcing --batch-size 1 square inference shape(1,3,{imgsz},{imgsz}) for non-PyTorch backends')
+
+ # Data
+ data = check_dataset(data) # check
+
+ # Configure
+ model.eval()
+ is_coco = isinstance(data.get('val'), str) and data['val'].endswith('coco/val2017.txt') # COCO dataset
+ nc = 1 if single_cls else int(data['nc']) # number of classes
+ iouv = torch.linspace(0.5, 0.95, 10).to(device) # iou vector for mAP@0.5:0.95
+ niou = iouv.numel()
+
+ names = {k: v for k, v in enumerate(model.names if hasattr(model, 'names') else model.module.names)}
+ # Dataloader
+ if not training:
+ model.warmup(imgsz=(1, 3, imgsz, imgsz), half=half) # warmup
+ pad = 0.0 if task == 'speed' else 0.5
+ task = task if task in ('train', 'val', 'test') else 'val' # path to train/val/test images
+ dataloader = create_dataloader(data[task], imgsz, batch_size, stride, names, single_cls, pad=pad, rect=pt,
+ workers=workers, prefix=colorstr(f'{task}: '))[0]
+
+ seen = 0
+ confusion_matrix = ConfusionMatrix(nc=nc)
+ # names = {k: v for k, v in enumerate(model.names if hasattr(model, 'names') else model.module.names)}
+ class_map = coco80_to_coco91_class() if is_coco else list(range(1000))
+ s = ('%20s' + '%11s' * 6) % ('Class', 'Images', 'Labels', 'P', 'R', 'HBBmAP@.5', ' HBBmAP@.5:.95')
+ dt, p, r, f1, mp, mr, map50, map = [0.0, 0.0, 0.0], 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0
+ # loss = torch.zeros(3, device=device)
+ loss = torch.zeros(4, device=device)
+ jdict, stats, ap, ap_class = [], [], [], []
+ pbar = tqdm(dataloader, desc=s, bar_format='{l_bar}{bar:10}{r_bar}{bar:-10b}') # progress bar
+ for batch_i, (im, targets, paths, shapes) in enumerate(pbar):
+ # targets (tensor): (n_gt_all_batch, [img_index clsid cx cy l s theta gaussian_θ_labels]) θ ∈ [-pi/2, pi/2)
+ # shapes (tensor): (b, [(h_raw, w_raw), (hw_ratios, wh_paddings)])
+ t1 = time_sync()
+ if pt or jit or engine:
+ im = im.to(device, non_blocking=True)
+ targets = targets.to(device)
+ im = im.half() if half else im.float() # uint8 to fp16/32
+ im /= 255 # 0 - 255 to 0.0 - 1.0
+ nb, _, height, width = im.shape # batch size, channels, height, width
+ t2 = time_sync()
+ dt[0] += t2 - t1
+
+ # Inference
+ out, train_out = model(im) if training else model(im, augment=augment, val=True) # inference, loss outputs
+ dt[1] += time_sync() - t2
+
+ # Loss
+ if compute_loss:
+ loss += compute_loss([x.float() for x in train_out], targets)[1] # box, obj, cls, theta
+
+ # NMS
+ # targets[:, 2:] *= torch.Tensor([width, height, width, height]).to(device) # to pixels
+ lb = [targets[targets[:, 0] == i, 1:] for i in range(nb)] if save_hybrid else [] # for autolabelling
+ t3 = time_sync()
+ # out = non_max_suppression(out, conf_thres, iou_thres, labels=lb, multi_label=True, agnostic=single_cls)
+ out = non_max_suppression_obb(out, conf_thres, iou_thres, labels=lb, multi_label=True, agnostic=single_cls) # list*(n, [xylsθ, conf, cls]) θ ∈ [-pi/2, pi/2)
+ dt[2] += time_sync() - t3
+
+ # Metrics
+ for si, pred in enumerate(out): # pred (tensor): (n, [xylsθ, conf, cls])
+ labels = targets[targets[:, 0] == si, 1:7] # labels (tensor):(n_gt, [clsid cx cy l s theta]) θ[-pi/2, pi/2)
+ nl = len(labels)
+ tcls = labels[:, 0].tolist() if nl else [] # target class
+ path, shape = Path(paths[si]), shapes[si][0] # shape (tensor): (h_raw, w_raw)
+ seen += 1
+
+ if len(pred) == 0:
+ if nl:
+ stats.append((torch.zeros(0, niou, dtype=torch.bool), torch.Tensor(), torch.Tensor(), tcls))
+ continue
+
+ # Predictions
+ if single_cls:
+ # pred[:, 5] = 0
+ pred[:, 6] = 0
+ poly = rbox2poly(pred[:, :5]) # (n, 8)
+ pred_poly = torch.cat((poly, pred[:, -2:]), dim=1) # (n, [poly, conf, cls])
+ hbbox = xywh2xyxy(poly2hbb(pred_poly[:, :8])) # (n, [x1 y1 x2 y2])
+ pred_hbb = torch.cat((hbbox, pred_poly[:, -2:]), dim=1) # (n, [xyxy, conf, cls])
+
+ pred_polyn = pred_poly.clone() # predn (tensor): (n, [poly, conf, cls])
+ scale_polys(im[si].shape[1:], pred_polyn[:, :8], shape, shapes[si][1]) # native-space pred
+ hbboxn = xywh2xyxy(poly2hbb(pred_polyn[:, :8])) # (n, [x1 y1 x2 y2])
+ pred_hbbn = torch.cat((hbboxn, pred_polyn[:, -2:]), dim=1) # (n, [xyxy, conf, cls]) native-space pred
+
+
+ # Evaluate
+ if nl:
+ # tbox = xywh2xyxy(labels[:, 1:5]) # target boxes
+ tpoly = rbox2poly(labels[:, 1:6]) # target poly
+ tbox = xywh2xyxy(poly2hbb(tpoly)) # target hbb boxes [xyxy]
+ scale_coords(im[si].shape[1:], tbox, shape, shapes[si][1]) # native-space labels
+ labels_hbbn = torch.cat((labels[:, 0:1], tbox), 1) # native-space labels (n, [cls xyxy])
+ correct = process_batch(pred_hbbn, labels_hbbn, iouv)
+ if plots:
+ confusion_matrix.process_batch(pred_hbbn, labels_hbbn)
+ else:
+ correct = torch.zeros(pred.shape[0], niou, dtype=torch.bool)
+ # stats.append((correct.cpu(), pred[:, 4].cpu(), pred[:, 5].cpu(), tcls)) # (correct, conf, pcls, tcls)
+ stats.append((correct.cpu(), pred_poly[:, 8].cpu(), pred_poly[:, 9].cpu(), tcls)) # (correct, conf, pcls, tcls)
+
+ # Save/log
+ if save_txt: # just save hbb pred results!
+ save_one_txt(pred_hbbn, save_conf, shape, file=save_dir / 'labels' / (path.stem + '.txt'))
+ # LOGGER.info('The horizontal prediction results has been saved in txt, which format is [cls cx cy w h /conf/]')
+ if save_json: # save hbb pred results and poly pred results.
+ save_one_json(pred_hbbn, pred_polyn, jdict, path, class_map) # append to COCO-JSON dictionary
+ # LOGGER.info('The hbb and obb results has been saved in json file')
+ callbacks.run('on_val_image_end', pred_hbb, pred_hbbn, path, names, im[si])
+
+ # Plot images
+ if plots and batch_i < 3:
+ f = save_dir / f'val_batch{batch_i}_labels.jpg' # labels
+ Thread(target=plot_images, args=(im, targets, paths, f, names), daemon=True).start()
+ f = save_dir / f'val_batch{batch_i}_pred.jpg' # predictions
+ Thread(target=plot_images, args=(im, output_to_target(out), paths, f, names), daemon=True).start()
+
+ # Compute metrics
+ stats = [np.concatenate(x, 0) for x in zip(*stats)] # to numpy
+ if len(stats) and stats[0].any():
+ tp, fp, p, r, f1, ap, ap_class = ap_per_class(*stats, plot=plots, save_dir=save_dir, names=names)
+ ap50, ap = ap[:, 0], ap.mean(1) # AP@0.5, AP@0.5:0.95
+ mp, mr, map50, map = p.mean(), r.mean(), ap50.mean(), ap.mean()
+ nt = np.bincount(stats[3].astype(np.int64), minlength=nc) # number of targets per class
+ else:
+ nt = torch.zeros(1)
+
+ # Print results
+ pf = '%20s' + '%11i' * 2 + '%11.3g' * 4 # print format
+ LOGGER.info(pf % ('all', seen, nt.sum(), mp, mr, map50, map))
+
+ # Print results per class
+ if (verbose or (nc < 50 and not training)) and nc > 1 and len(stats):
+ for i, c in enumerate(ap_class):
+ LOGGER.info(pf % (names[c], seen, nt[c], p[i], r[i], ap50[i], ap[i]))
+
+ # Print speeds
+ t = tuple(x / seen * 1E3 for x in dt) # speeds per image
+ if not training:
+ shape = (batch_size, 3, imgsz, imgsz)
+ LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {shape}' % t)
+
+ # Plots
+ if plots:
+ confusion_matrix.plot(save_dir=save_dir, names=list(names.values()))
+ callbacks.run('on_val_end')
+
+ # Save JSON
+ if save_json and len(jdict):
+ w = Path(weights[0] if isinstance(weights, list) else weights).stem if weights is not None else '' # weights
+ pred_json = str(save_dir / f"{w}_obb_predictions.json") # predictions json
+ LOGGER.info(f'\nEvaluating pycocotools mAP... saving {pred_json}...')
+ with open(pred_json, 'w') as f:
+ json.dump(jdict, f)
+ LOGGER.info('---------------------The hbb and obb results has been saved in json file-----------------------')
+ anno_json = str(Path(data.get('path', '../coco')) / 'annotations/instances_val2017.json') # annotations json
+ try: # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
+ check_requirements(['pycocotools'])
+ from pycocotools.coco import COCO
+ from pycocotools.cocoeval import COCOeval
+
+ anno = COCO(anno_json) # init annotations api
+ pred = anno.loadRes(pred_json) # init predictions api
+ eval = COCOeval(anno, pred, 'bbox')
+ if is_coco:
+ eval.params.imgIds = [int(Path(x).stem) for x in dataloader.dataset.img_files] # image IDs to evaluate
+ eval.evaluate()
+ eval.accumulate()
+ eval.summarize()
+ map, map50 = eval.stats[:2] # update results (mAP@0.5:0.95, mAP@0.5)
+ except Exception as e:
+ LOGGER.info(f'pycocotools unable to run: {e}')
+
+ # Return results
+ model.float() # for training
+ if not training:
+ s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
+ LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
+ maps = np.zeros(nc) + map
+ for i, c in enumerate(ap_class):
+ maps[c] = ap[i]
+ return (mp, mr, map50, map, *(loss.cpu() / len(dataloader)).tolist()), maps, t
+
+
+def parse_opt():
+ parser = argparse.ArgumentParser()
+ parser.add_argument('--data', type=str, default=ROOT / 'data/HRSC2016.yaml', help='dataset.yaml path')
+ parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'runs/train/exp30/weights/last.pt', help='model.pt path(s)')
+ parser.add_argument('--batch-size', type=int, default=4, help='batch size')
+ parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=768, help='inference size (pixels)')
+ parser.add_argument('--conf-thres', type=float, default=0.01, help='confidence threshold')
+ parser.add_argument('--iou-thres', type=float, default=0.4, help='NMS IoU threshold')
+ parser.add_argument('--task', default='val', help='train, val, test, speed or study')
+ parser.add_argument('--device', default='0', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+ parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
+ parser.add_argument('--single-cls', action='store_true', help='treat as single-class dataset')
+ parser.add_argument('--augment', action='store_true', help='augmented inference')
+ parser.add_argument('--verbose', action='store_true', help='report mAP by class')
+ parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
+ parser.add_argument('--save-hybrid', action='store_true', help='save label+prediction hybrid results to *.txt')
+ parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
+ parser.add_argument('--save-json', action='store_true', help='save a COCO-JSON results file')
+ parser.add_argument('--project', default=ROOT / 'runs/val', help='save to project/name')
+ parser.add_argument('--name', default='exp', help='save to project/name')
+ parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
+ parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
+ parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
+ opt = parser.parse_args()
+ opt.data = check_yaml(opt.data) # check YAML
+ opt.save_json |= opt.data.endswith('coco.yaml')
+ opt.save_txt |= opt.save_hybrid
+ print_args(FILE.stem, opt)
+ return opt
+
+
+def main(opt):
+ check_requirements(requirements=ROOT / 'requirements.txt', exclude=('tensorboard', 'thop'))
+
+ if opt.task in ('train', 'val', 'test'): # run normally
+ # if opt.conf_thres > 0.001: # https://github.com/ultralytics/yolov5/issues/1466
+ if opt.conf_thres > 0.01:
+ LOGGER.info(f'WARNING: In oriented detection, confidence threshold {opt.conf_thres} >> 0.01 will produce invalid mAP values.')
+ run(**vars(opt))
+
+ else:
+ weights = opt.weights if isinstance(opt.weights, list) else [opt.weights]
+ opt.half = True # FP16 for fastest results
+ if opt.task == 'speed': # speed benchmarks
+ # python val.py --task speed --data coco.yaml --batch 1 --weights yolov5n.pt yolov5s.pt...
+ opt.conf_thres, opt.iou_thres, opt.save_json = 0.25, 0.45, False
+ for opt.weights in weights:
+ run(**vars(opt), plots=False)
+
+ elif opt.task == 'study': # speed vs mAP benchmarks
+ # python val.py --task study --data coco.yaml --iou 0.7 --weights yolov5n.pt yolov5s.pt...
+ for opt.weights in weights:
+ f = f'study_{Path(opt.data).stem}_{Path(opt.weights).stem}.txt' # filename to save to
+ x, y = list(range(256, 1536 + 128, 128)), [] # x axis (image sizes), y axis
+ for opt.imgsz in x: # img-size
+ LOGGER.info(f'\nRunning {f} --imgsz {opt.imgsz}...')
+ r, _, t = run(**vars(opt), plots=False)
+ y.append(r + t) # results and times
+ np.savetxt(f, y, fmt='%10.4g') # save
+ os.system('zip -r study.zip study_*.txt')
+ plot_val_study(x=x) # plot
+
+
+if __name__ == "__main__":
+ opt = parse_opt()
+ main(opt)