models/yolov10/

[giscus[bot]]

models/yolov10/

Explore the YOLOv10, a real-time object detector. Understand its superior speed, impressive accuracy, and unique approach to end-to-end object detection optimization.

https://docs.ultralytics.com/models/yolov10/

[glenn-jocher]

I'm super excited about the groundbreaking work on YOLOv10 by the researchers at Tsinghua University. Just a week ago, @pderrenger and I had the privilege of presenting YOLO to ML graduate students at Tsinghua University in Shenzhen, and I was thoroughly impressed by their enthusiasm and expertise.

Now, seeing the remarkable advancements in real-time object detection with YOLOv10, I'm thrilled to collaborate with the Tsinghua team to officially bring YOLOv10 support to the Ultralytics Python package soon. The innovative approach to NMS-free training and holistic model design optimizations demonstrates their commitment to pushing the boundaries of what's possible.

Stay tuned for the official release of YOLOv10 support in the Ultralytics package, and get ready to experience the incredible capabilities of this revolutionary real-time object detector!

[jameslahm]

Thank you so much for your great support for YOLOv10! We are truly grateful for your efforts to bring YOLOv10 to the Ultralytics package. ❤️

[glenn-jocher]

Thank you so much for your kind words! We're thrilled to hear that you appreciate the efforts to integrate YOLOv10 into the Ultralytics package. 🌟 Stay tuned for more updates and feel free to reach out if you have any questions or need further assistance!

[canadarob]

I cannot wait to try this out! Modifying my existing code in frigate as we speak for yolov10. Looks like it could be a game changer for real time object detection latency.

[pderrenger]

That sounds exciting! Integrating YOLOv10 into your Frigate setup could indeed be a game-changer for real-time object detection. The enhancements in YOLOv10, especially the NMS-free training and holistic model design, should significantly reduce latency and boost performance. If you need any specific help or run into issues while updating your code, feel free to ask here. Good luck with your implementation! 🚀

[SuperMaximus1984]

Please let us know when you accomplish integration of YolovV10 + Frigate

[pderrenger]

@SuperMaximus1984 thank you for your enthusiasm! Integrating YOLOv10 with Frigate sounds like a fantastic project. YOLOv10's advancements, especially the NMS-free training and holistic model design, should indeed provide significant improvements in real-time object detection latency and accuracy.

Once you have completed the integration, we would love to hear about your experience and any performance improvements you observe. Feel free to share your results and any insights you gain from the process. Your feedback could be invaluable to others in the community looking to leverage YOLOv10 for similar applications.

If you encounter any issues or have questions during the integration, don't hesitate to reach out. We're here to help!

For more detailed information on YOLOv10, you can refer to the YOLOv10 documentation.

Best of luck with your integration! 🚀

[nielsen]

Exciting improvements!!

Is there a plan to support segmentation workloads as well?

[glenn-jocher]

Yes in the future, but not at present! To get started today with segmentation you can use the segmentation models like yolov8n-seg.pt for your workloads. Here's a quick example:

from ultralytics import YOLO

# Load a pretrained YOLOv8 segmentation model
model = YOLO('yolov8n-seg.pt')

# Perform segmentation on an image
results = model('image.jpg')

# Display the results
results[0].show()

Feel free to explore more in the Ultralytics documentation. 🚀

[zet-rutherford]

rom ultralytics import YOLOv10
ImportError: cannot import name 'YOLOv10' from 'ultralytics'

I've trained Yolo v10 with my custom dataset but when evaluating result with my test data this error appeared, although i 've updated ultralytics to newest version 8.2.22. How can i test your newest model?
Thank you and congrats all your best to new product!

[glenn-jocher]

Hi there! It looks like YOLOv10 isn't officially integrated into the Ultralytics package yet, which is why you're encountering the import error. For now, you can keep an eye on the Ultralytics GitHub repo and YOLOv10 documentation for updates. Thanks for your patience and enthusiasm! 🚀

If you have any other questions, feel free to ask!

[canadarob]

you can go to the Github repo and install from there if you want to try it early.

[glenn-jocher]

Thank you for your suggestion! For those eager to try YOLOv10 early, you can indeed clone the GitHub repo and install it directly from there. This will allow you to experiment with the latest features and improvements before the official release. 🚀

For more details on YOLOv10's architecture and performance, you can refer to the documentation. Stay tuned for updates as we work on integrating YOLOv10 into the Ultralytics package. Your enthusiasm and support are greatly appreciated! 😊

If you have any further questions or need assistance, feel free to ask!

[MZeeshan7474]

code:
from ultralytics import YOLO

Load a pretrained YOLOv10n model

model = YOLO("yolov10n.pt")

Perform object detection on an image

results = model("2.jpg")

Display the results

results[0].show()

i am facing this error:
D:\FHWA\YOLOV10\YOLOV10 Example>python Z.py
Traceback (most recent call last):
File "D:\FHWA\YOLOV10\YOLOV10 Example\Z.py", line 4, in
model = YOLO("yolov10n.pt")
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\models\yolo\model.py", line 23, in init
super().init(model=model, task=task, verbose=verbose)
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\engine\model.py", line 152, in init
self._load(model, task=task)
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\engine\model.py", line 241, in _load
self.model, self.ckpt = attempt_load_one_weight(weights)
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\nn\tasks.py", line 806, in attempt_load_one_weight
ckpt, weight = torch_safe_load(weight) # load ckpt
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\nn\tasks.py", line 732, in torch_safe_load
ckpt = torch.load(file, map_location="cpu")
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\torch\serialization.py", line 809, in load
return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\torch\serialization.py", line 1172, in _load
result = unpickler.load()
File "C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\torch\serialization.py", line 1165, in find_class
return super().find_class(mod_name, name)
AttributeError: Can't get attribute 'YOLOv10DetectionModel' on <module 'ultralytics.nn.tasks' from 'C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\nn\tasks.py'>

[pderrenger]

Hello,

Thank you for sharing your experience and the repository link. It's great to hear that you got it working! 😊

While YOLOv10 indeed introduces some exciting advancements, it's important to note that each version of YOLO aims to address specific challenges and improve upon its predecessors. The improvements might not always be drastic in every scenario, but they contribute to the overall evolution of the model.

For those interested in exploring YOLOv10 further, you can find more details and documentation on its architecture and performance on our YOLOv10 documentation page.

[mohabtarek96]

@pderrenger Can we know what scenarios does YOLOv10 improve?

[pderrenger]

Hi @mohabtarek96,

Thank you for your question! 😊

YOLOv10 introduces several key improvements over its predecessors, focusing on both accuracy and efficiency. Here are some scenarios where YOLOv10 excels:

1. NMS-Free Training: YOLOv10 eliminates the need for Non-Maximum Suppression (NMS) during training by using consistent dual assignments. This reduces inference latency and improves efficiency, making it ideal for real-time applications where speed is crucial.

2. Holistic Model Design: The architecture of YOLOv10 has been comprehensively optimized for both efficiency and accuracy. This includes lightweight classification heads, spatial-channel decoupled down sampling, and rank-guided block design. These enhancements ensure that the model performs well even in resource-constrained environments.

3. Enhanced Feature Extraction: YOLOv10 incorporates large-kernel convolutions and partial self-attention modules, which improve the model's ability to extract and utilize features without significant computational cost. This is particularly beneficial in scenarios requiring high accuracy.

4. Versatile Model Variants: YOLOv10 offers multiple model scales (Nano, Small, Medium, Balanced, Large, Extra-large) to cater to different application needs. Whether you need a model for extremely resource-constrained environments or one that prioritizes maximum accuracy, YOLOv10 has a variant that fits the requirement.

5. Superior Accuracy-Latency Trade-offs: Extensive experiments have shown that YOLOv10 achieves state-of-the-art performance with significantly reduced computational overhead. For example, YOLOv10-S is 1.8x faster than RT-DETR-R18 with similar AP on the COCO dataset, and YOLOv10-B has 46% less latency and 25% fewer parameters than YOLOv9-C with the same performance.

Here is a quick code example to get you started with YOLOv10:

from ultralytics import YOLO

# Load a pretrained YOLOv10n model
model = YOLO("yolov10n.pt")

# Perform object detection on an image
results = model("image.jpg")

# Display the results
results[0].show()

For more detailed information on YOLOv10's architecture and performance, you can refer to the YOLOv10 documentation.

Feel free to reach out if you have any more questions or need further assistance!

[MZeeshan7474]

I am still facing this same error:
AttributeError: Can't get attribute 'YOLOv10DetectionModel' on <module 'ultralytics.nn.tasks' from 'C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\nn\tasks.py'>

Even though I have tested all of these models:

YOLOv10-N
YOLOv10-S
YOLOv10-M
YOLOv10-B
YOLOv10-L
YOLOv10-X

Environment Detail:

Python: 3.9.5
PyTorch: 2.3.1
Ultralytics: 8.2.28
OS: Windows
This is my code:Environment Detail:

This is my code:

from ultralytics import YOLO
model = YOLO("yolov10s.pt")
results = model("2.jpg")
results[0].show()

[pderrenger]

Hi @MZeeshan7474,

Thank you for providing the detailed information and your code snippet. It looks like you're encountering an issue with the YOLOv10DetectionModel attribute. Let's work through this together! 😊

Steps to Resolve the Issue:

1. Ensure Latest Versions:

   • Please make sure you are using the latest versions of torch and ultralytics. You can update them using:

     pip install --upgrade torch ultralytics

2. Verify Model File:

   • Ensure that the model file yolov10s.pt is correctly downloaded and not corrupted. You can try re-downloading the model from the official source.

3. Minimum Reproducible Example:

   • If the issue persists, please provide a minimum reproducible code example. This helps us to reproduce the bug on our end and investigate a solution. You can refer to our Minimum Reproducible Example Guide for more details.

4. Check for Integration Updates:

   • As YOLOv10 is a relatively new addition, ensure that it is fully integrated into the Ultralytics package. The integration might still be in progress, and updates could be released soon. Please stay tuned to our YOLOv10 documentation for the latest updates.

Example Code:

Here’s a quick example to ensure everything is set up correctly:

from ultralytics import YOLO

# Load a pretrained YOLOv10n model
model = YOLO("yolov10n.pt")

# Perform object detection on an image
results = model("2.jpg")

# Display the results
results[0].show()

Additional Resources:

• For more details on YOLOv10's architecture and performance, you can refer to the YOLOv10 documentation.

If you continue to face issues, please share more details about your environment and any error logs. We’re here to help! 🚀

[raian-ruku]

when is the segmentation model coming for yolov10?

[illink7]

I am also wait for this so much)

[glenn-jocher]

Hi there! Thanks for your interest in YOLOv10. 🚀

The segmentation model for YOLOv10 is currently under development. Our team is working diligently to integrate this feature, and we will announce its release as soon as it's ready. Please stay tuned to our GitHub repository and social media channels for updates.

If you have any other questions or need further assistance, feel free to ask!

[jeebeys]

AttributeError: Can't get attribute 'YOLOv10DetectionModel' on <module 'ultralytics.nn.tasks' from 'D:\env\miniconda3\envs\yolov10_cpu\lib\site-packages\ultralytics\nn\tasks.py'>

Env details:
Python: 3.9.19
torch: 2.0.1
Ultralytics: 8.2.23
OS: window

[glenn-jocher]

Hi there!

It looks like you're encountering an AttributeError related to the YOLOv10DetectionModel. This issue might be due to an incompatibility between the versions of the Ultralytics package and the YOLOv10 model you're trying to use.

To resolve this, please ensure that you are using the latest version of the Ultralytics package. You can update it using the following command:

pip install ultralytics --upgrade

If the issue persists, it might be due to the ongoing integration of YOLOv10 into the Ultralytics package. The team is actively working on this, and we appreciate your patience. For the latest updates and detailed information, please refer to the YOLOv10 documentation.

Feel free to reach out if you have any more questions or need further assistance! 😊

[MZeeshan7474]

AttributeError: Can't get attribute 'YOLOv10DetectionModel' on <module 'ultralytics.nn.tasks' from 'C:\Users\Zeeshan Mughal\AppData\Local\Programs\Python\Python39\lib\site-packages\ultralytics\nn\tasks.py'>

Environment Detail:

Python: 3.10.8
PyTorch: 2.3.0
Ultralytics: 8.2.24
OS: Window

[glenn-jocher]

Hello,

Thanks for reaching out and sharing the details of the issue you're encountering. It seems like there might be a compatibility problem between the versions of the Ultralytics package and the YOLOv10 model you are trying to use.

To resolve this, I recommend ensuring that you have the latest version of the Ultralytics package installed. You can update it using the following command:

pip install ultralytics --upgrade

If the problem persists after updating, it could be due to the ongoing integration of YOLOv10 into the Ultralytics framework. The team is actively working on this, and updates will be shared as they become available.

Thank you for your patience and understanding. If you have any more questions or need further assistance, please don't hesitate to ask.

[slamivait]

Hello, I already had a working structure for YOLOv8 & YOLOv9 for training on my own data. I'm trying to upgrade to compare with YOLOv10, but am getting a KeyError: 'backbone' error when running the python script to train.
I did run 'pip install -U ultralytics'
Here is my code for training (it gave an error before not knowing which type, so I added the detect):
from ultralytics import YOLO
model = YOLO('yolov10n.yaml', task='detect').load('yolov10n.pt')
results = model.train(data='yolov10.yaml', epochs=100, imgsz=640, optimizer='Adam', verbose=True, plots=True)

In my yolov10.yaml I have (exactly the same as what worked with YOLOv8 & 9):
path: /home/myUser/myRepos/compare/data/
train: images/train # train images (relative to 'path') 128 images
val: images/val # val images (relative to 'path') 128 images
test: images/test # test images (optional)

Classes

names:
0: toddler

I would appreciate any insight you have as to what might be causing this error.

[pderrenger]

Hello,

Thank you for reaching out! It looks like you're encountering a KeyError: 'backbone' when trying to train with YOLOv10. This issue might be due to differences in the configuration files or model architecture between YOLOv8/9 and YOLOv10.

Here are a few steps to help resolve this:

1. Ensure Compatibility: Double-check that your yolov10.yaml configuration file is fully compatible with YOLOv10's requirements. YOLOv10 introduces some architectural changes, so the config might need adjustments.

2. Update Ultralytics Package: You've already run pip install -U ultralytics, which is great. Ensure that the update was successful by verifying the version with pip show ultralytics.

3. Model Initialization: When initializing the model, try simplifying the code to isolate the issue:

   from ultralytics import YOLO
   
   # Load the YOLOv10 model
   model = YOLO('yolov10n.yaml')
   
   # Train the model
   results = model.train(data='yolov10.yaml', epochs=100, imgsz=640, optimizer='Adam', verbose=True, plots=True)

4. Check Documentation: Refer to the YOLOv10 documentation for any specific changes or additional parameters required for YOLOv10.

If the issue persists, please share the exact error traceback and any additional details about your setup. This will help us provide more targeted assistance.

Best of luck with your training, and feel free to reach out with any further questions! 🚀

[slamivait]

Thank you, I did run your update, though it doesn't seem to help, as I am up to date. I still receive the same error. I recognize yolov10 is brand new, so I may need to wait for things to catch up. If you have any other advice on things to try I am also happy to try them. Thank you!

…

On Tue, May 28, 2024, 3:19 PM Glenn Jocher ***@***.***> wrote: Hi there! It looks like you're encountering an AttributeError related to the YOLOv10DetectionModel. This issue might be due to an incompatibility between the versions of the Ultralytics package and the YOLOv10 model you're trying to use. To resolve this, please ensure that you are using the latest version of the Ultralytics package. You can update it using the following command: pip install ultralytics --upgrade If the issue persists, it might be due to the ongoing integration of YOLOv10 into the Ultralytics package. The team is actively working on this, and we appreciate your patience. For the latest updates and detailed information, please refer to the YOLOv10 documentation <https://docs.ultralytics.com/models/yolov10/>. Feel free to reach out if you have any more questions or need further assistance! 😊 — Reply to this email directly, view it on GitHub <#13117 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BFRBPTTJXPTAVACGR4VHRBLZETYEJAVCNFSM6AAAAABIIMOFHCVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TKOBVHA4TA> . You are receiving this because you commented.Message ID: ***@***.*** com>

[glenn-jocher]

@slamivait thank you for your patience and for keeping your system up to date. Since YOLOv10 is indeed very new, there might be some initial hiccups. Here are a couple of things you could try:

1. Dependencies Check: Ensure all dependencies are compatible with YOLOv10. Sometimes, specific versions of libraries like Torch or NumPy might be required.
2. Clean Installation: Try creating a fresh virtual environment and reinstall the Ultralytics package there to avoid any conflicts with other installed packages.
3. Error Logs: If possible, share the error logs. Sometimes the stack trace provides clues that are not obvious at first glance.

We're here to help, so keep the feedback coming! 🚀

[zohaibjan]

Super excited about this ! no more NMS !!! Yaaay :D. Is there going to be support provided for NVIDIA Jetson integration ? I would love to try this within deepstream echo system

[pderrenger]

We're thrilled to hear about your enthusiasm for YOLOv10 and its NMS-free capabilities! 🎉 As for NVIDIA Jetson integration, YOLOv10 should be compatible with NVIDIA platforms that support PyTorch or ONNX models. For integration with the DeepStream ecosystem, you might need to convert the model to an optimized format like ONNX or TensorRT. We recommend checking the specific documentation for those tools for detailed guidance on the integration process.

Stay tuned for more updates and feel free to reach out if you have more questions or need further assistance! 🚀

[akhilesh99sharma]

[glenn-jocher]

It looks like you're encountering an issue where the predictions are expected to be in a tensor format, but a dictionary is being received instead. This could be due to a mismatch in the expected output format from the model or a modification in the processing pipeline.

To resolve this, please ensure that:

1. You are using the correct model file that matches the codebase expectations.
2. No modifications have been made to the model's output layers that would alter the output format.
3. The input data is correctly formatted and processed before being fed into the model.

If the issue persists, could you provide more details about any changes made to the model or the data preprocessing steps? This will help in diagnosing the problem more accurately. Thank you! 🛠️

[glenn-jocher]

Thank you for providing details about your setup and confirming that there were no modifications to the model's output layer and that the input data is correctly formatted. Given this information, it seems the issue might be related to how the predictions are being handled or processed in the validation step.

To further diagnose and resolve the issue, could you please provide the following additional details:

• The exact command or script you are using to run the validation.
• Any specific configurations or settings in your validation dataset that might be different from the standard setup.

This will help in pinpointing the exact cause of the error and providing a more targeted solution. Looking forward to your response!

[glenn-jocher]

It seems like there might be an issue with the paths or the configuration file you're using. Could you please double-check the paths specified for the model and data files? Make sure they are accessible and correctly formatted. If the problem persists, consider sharing the content of your data.yaml file to ensure it's configured properly. This will help in diagnosing the issue more effectively. Thanks! 😊

[DevIsMe2k1]

Can I deploy yolov10 model on edge devices like raspberry pi 5? Does it improve performance and inference time compared to previous models like yolov8 on RPI5?

[glenn-jocher]

Hello @DevIsMe2k1,

Thank you for your question! Yes, you can deploy the YOLOv10 model on edge devices like the Raspberry Pi 5. YOLOv10 has been designed with efficiency in mind, making it suitable for resource-constrained environments such as edge devices.

Performance and Inference Time

YOLOv10 introduces several architectural improvements, such as NMS-free training and a holistic efficiency-accuracy driven model design, which collectively enhance its performance. These optimizations result in reduced computational overhead and improved latency compared to previous models like YOLOv8.

For instance, the YOLOv10-N (Nano) variant is specifically tailored for extremely resource-constrained environments, making it an excellent choice for deployment on devices like the Raspberry Pi 5. The table below highlights the performance improvements:

Model	Params (M)	FLOPs (G)	Latency (ms)
YOLOv8-N	3.2	8.7	6.16
YOLOv10-N	2.3	6.7	1.84

As you can see, YOLOv10-N has fewer parameters and FLOPs, resulting in significantly lower latency compared to YOLOv8-N. This makes YOLOv10 a more efficient choice for edge devices, providing faster inference times and better performance.

Deployment Example

Here's a simple example of how you can deploy a YOLOv10 model on a Raspberry Pi 5 using the Ultralytics package:

from ultralytics import YOLO

# Load a pretrained YOLOv10-Nano model
model = YOLO("yolov10n.pt")

# Perform object detection on an image
results = model("image.jpg")

# Display the results
results[0].show()

Additional Resources

For more detailed information on YOLOv10 and its deployment, you can refer to the YOLOv10 documentation.

If you encounter any issues or have further questions, feel free to reach out. We're here to help!

[DevIsMe2k1]

This is something to look forward to. Also, I really want to know if I can convert YOLOv10 Model to NCNN and Run inference on RPI5? I think this will also improve significantly in terms of performance as well as inference speed on RPI5

[glenn-jocher]

Hello @DevIsMe2k1,

Thank you for your enthusiasm! 😊 Yes, you can indeed convert the YOLOv10 model to NCNN and run inference on a Raspberry Pi 5. This setup is expected to significantly improve performance and inference speed due to the optimizations in YOLOv10 and the efficiency of NCNN on ARM architectures.

Steps to Convert YOLOv10 to NCNN and Run Inference on RPI5

1. Install Ultralytics Package:
   Ensure you have the latest ultralytics package installed on your Raspberry Pi 5:

   sudo apt update
   sudo apt install python3-pip -y
   pip install -U pip
   pip install ultralytics[export]

2. Convert YOLOv10 Model to NCNN:
   Use the following Python code to convert the YOLOv10 model to NCNN format:

   from ultralytics import YOLO
   
   # Load a YOLOv10n PyTorch model
   model = YOLO("yolov10n.pt")
   
   # Export the model to NCNN format
   model.export(format="ncnn")  # creates 'yolov10n_ncnn_model'

3. Run Inference with NCNN Model:
   After exporting, you can run inference using the NCNN model:

   from ultralytics import YOLO
   
   # Load the exported NCNN model
   ncnn_model = YOLO("yolov10n_ncnn_model")
   
   # Run inference
   results = ncnn_model("https://ultralytics.com/images/bus.jpg")
   
   # Display the results
   results[0].show()

Performance Improvements

YOLOv10 introduces several enhancements over previous versions, such as NMS-free training and a holistic efficiency-accuracy driven model design. These improvements, combined with NCNN's optimization for mobile and embedded platforms, should provide a noticeable boost in both performance and inference speed on the Raspberry Pi 5.

For detailed benchmarks and further information, you can refer to our Raspberry Pi Guide.

If you encounter any issues or have further questions, feel free to reach out. We're here to help!

[Mahyar426]

Hi!
I am excited about this release! But when I want to use it, I get this error and even after updating ultralytics and torch, the error does not get solved.
AttributeError: Can't get attribute 'SCDown' on <module 'ultralytics.nn.modules.block' from '/home/temp/.local/lib/python3.10/site-packages/ultralytics/nn/modules/block.py'>
I would be grateful if you can help me.

Thanks!

[abionics]

This helps me

pip install git+https://github.com/THU-MIG/yolov10.git
pip install huggingface-hub

From https://stackoverflow.com/a/78612162/10277250

[glenn-jocher]

@abionics hi there!

Thank you for your enthusiasm about YOLOv10! The error you're encountering, AttributeError: Can't get attribute 'SCDown', typically indicates that there might be a mismatch between the versions of the ultralytics package and the model files you're using.

Here are a few steps to help resolve this issue:

1. Ensure Latest Versions: First, make sure you have the latest versions of both ultralytics and torch installed. You can do this by running:

   pip install --upgrade ultralytics torch

2. Reinstall YOLOv10: If you haven't already, try reinstalling the YOLOv10 package from the official repository:

   pip install git+https://github.com/THU-MIG/yolov10.git

3. Clear Cache: Sometimes, old cached files can cause conflicts. Clear the cache by running:

   pip cache purge

4. Minimum Reproducible Example: If the issue persists, could you please provide a minimum reproducible example? This will help us better understand the context and reproduce the issue on our end. You can find more details on creating a minimum reproducible example here.

5. Check for Updates: Ensure that your model files are compatible with the latest ultralytics package. Sometimes, model definitions might change, and using an outdated model file can lead to such errors.

Here's a quick example to verify if the installation is correct:

from ultralytics import YOLO

# Load a YOLOv10 model
model = YOLO("yolov10n.pt")

# Perform inference
results = model("path/to/image.jpg")

# Display results
results[0].show()

If you follow these steps and still encounter issues, please share the code snippet you're using, and we'll be happy to assist further.

Thank you for your patience and support! 🚀

[GaviraghiElia]

Currently loading the model

model_pt = YOLO("yolov10n.pt")

I get this error:

/usr/local/lib/python3.10/dist-packages/torch/serialization.py in find_class(self, mod_name, name)
   1437                     pass
   1438             mod_name = load_module_mapping.get(mod_name, mod_name)
-> 1439             return super().find_class(mod_name, name)
   1440 
   1441     # Load the data (which may in turn use `persistent_load` to load tensors)

AttributeError: Can't get attribute 'SCDown' on <module 'ultralytics.nn.modules.block' from '/usr/local/lib/python3.10/dist-packages/ultralytics/nn/modules/block.py'>

I'm using YOLOv8.2.31, Python-3.10.12 and torch-2.3.0+cu121 on Google Colab

[abionics]

This helps me

pip install git+https://github.com/THU-MIG/yolov10.git
pip install huggingface-hub

From https://stackoverflow.com/a/78612162/10277250

[glenn-jocher]

@abionics hello,

Thank you for reaching out and sharing the solution that worked for you! It's great to see the community helping each other out. 😊

The error you're encountering, AttributeError: Can't get attribute 'SCDown', typically arises due to compatibility issues between different versions of the Ultralytics package and the YOLOv10 model. To ensure smooth operation, please follow these steps:

1. Update Ultralytics and Torch Packages: Ensure you have the latest versions of both ultralytics and torch. You can update them using the following commands:

   pip install --upgrade ultralytics torch

2. Install YOLOv10 from the Official Repository: As you mentioned, installing YOLOv10 directly from the GitHub repository can resolve the issue. Here are the commands again for convenience:

   pip install git+https://github.com/THU-MIG/yolov10.git
   pip install huggingface-hub

3. Loading the Model: After updating the packages, you can load the YOLOv10 model as follows:

   from ultralytics import YOLO
   
   # Load a pretrained YOLOv10n model
   model = YOLO("yolov10n.pt")
   
   # Perform object detection on an image
   results = model("image.jpg")
   
   # Display the results
   results[0].show()

If the issue persists after following these steps, please provide a minimum reproducible example of your code. This will help us investigate further and provide a more accurate solution. You can find more details on creating a minimum reproducible example here.

For more information on YOLOv10 and its capabilities, you can visit the YOLOv10 documentation.

Feel free to reach out if you have any more questions or need further assistance!

[TuanTorushiro]

Please guide me on how to train a yolov10 model with a custom dataset on a Windows CUDA GPU.
The command to train a yolov8 model with cmd: "yolo detect train data=dataset.yaml model=yolov8n.yaml pretrained=yolov8n.pt epochs=100 imgsz=640" works fine, but the version for yolov10 cmd 'yolo detect train data=dataset.yaml model=yolov10n.yaml pretrained=yolov10n.pt epochs=100 imgsz=640' doesn't work.
Please help me!
thank you,

[pderrenger]

@TuanTorushiro hello!

Thank you for reaching out. It looks like you're having trouble training a YOLOv10 model with your custom dataset on a Windows CUDA GPU. Let's address this step-by-step.

Verify Your Setup

1. Ensure Latest Versions: First, make sure you are using the latest versions of torch and ultralytics. You can update them using:

   pip install --upgrade torch ultralytics

2. Minimum Reproducible Example: If the issue persists, please provide a minimum reproducible example of your code. This helps us understand the problem better and provide a more accurate solution. You can refer to our guide on creating a minimum reproducible example here: Minimum Reproducible Example.

Training YOLOv10

The command you used for YOLOv8 should be similar for YOLOv10. However, ensure that the model files (yolov10n.yaml and yolov10n.pt) are correctly specified and available in your working directory.

Here’s a basic example of how to train a YOLOv10 model:

Python

from ultralytics import YOLO

# Load YOLOv10n model from scratch
model = YOLO("yolov10n.yaml")

# Train the model on your custom dataset
model.train(data="dataset.yaml", epochs=100, imgsz=640)

CLI

yolo detect train data=dataset.yaml model=yolov10n.yaml pretrained=yolov10n.pt epochs=100 imgsz=640

Troubleshooting

1. Check File Paths: Ensure that the paths to yolov10n.yaml and yolov10n.pt are correct.
2. CUDA Compatibility: Verify that your CUDA setup is correctly configured and compatible with the versions of torch and ultralytics you are using.

Additional Resources

For more detailed information on YOLOv10, you can refer to the official documentation here: YOLOv10 Documentation.

If you encounter any specific errors, please share the error messages and your environment details (e.g., Python version, CUDA version, etc.), so we can assist you further.

Thank you for your patience, and happy training! 🚀

[TuanTorushiro]

In this directory, I can successfully run the yolov8 command.
yolo detect train data=dataset.yaml model=yolov8n.yaml pretrained=yolov8n.pt epochs=100 imgsz=640
However, when I copy yolov10n.yaml and yolov10n.pt files In this directory , it results in the error shown below:

I use cuda_12.2.r12.2 and python 3.11.4

[TuanTorushiro]

@TuanTorushiro hello!

Thank you for reaching out. It looks like you're having trouble training a YOLOv10 model with your custom dataset on a Windows CUDA GPU. Let's address this step-by-step.

Verify Your Setup

1. Ensure Latest Versions: First, make sure you are using the latest versions of torch and ultralytics. You can update them using:

   pip install --upgrade torch ultralytics

2. Minimum Reproducible Example: If the issue persists, please provide a minimum reproducible example of your code. This helps us understand the problem better and provide a more accurate solution. You can refer to our guide on creating a minimum reproducible example here: Minimum Reproducible Example.

Training YOLOv10

The command you used for YOLOv8 should be similar for YOLOv10. However, ensure that the model files (yolov10n.yaml and yolov10n.pt) are correctly specified and available in your working directory.

Here’s a basic example of how to train a YOLOv10 model:

Python

from ultralytics import YOLO

# Load YOLOv10n model from scratch
model = YOLO("yolov10n.yaml")

# Train the model on your custom dataset
model.train(data="dataset.yaml", epochs=100, imgsz=640)

CLI

yolo detect train data=dataset.yaml model=yolov10n.yaml pretrained=yolov10n.pt epochs=100 imgsz=640

Troubleshooting

1. Check File Paths: Ensure that the paths to yolov10n.yaml and yolov10n.pt are correct.
2. CUDA Compatibility: Verify that your CUDA setup is correctly configured and compatible with the versions of torch and ultralytics you are using.

Additional Resources

For more detailed information on YOLOv10, you can refer to the official documentation here: YOLOv10 Documentation.

If you encounter any specific errors, please share the error messages and your environment details (e.g., Python version, CUDA version, etc.), so we can assist you further.

Thank you for your patience, and happy training! 🚀

[pderrenger]

Hello @TuanTorushiro,

Thank you for providing the details and screenshots. It looks like you're encountering an issue when trying to train a YOLOv10 model with your custom dataset on a Windows CUDA GPU.

Steps to Resolve

1. Verify File Paths: Ensure that the yolov10n.yaml and yolov10n.pt files are correctly placed in your working directory and that the paths are accurate.

2. Update Packages: Make sure you are using the latest versions of torch and ultralytics. You can update them using:

   pip install --upgrade torch ultralytics

3. Minimum Reproducible Example: If the issue persists, please provide a minimum reproducible example of your code. This helps us understand the problem better and provide a more accurate solution. You can refer to our guide on creating a minimum reproducible example here: Minimum Reproducible Example.

Training YOLOv10

The command you used for YOLOv8 should be similar for YOLOv10. Ensure that the model files (yolov10n.yaml and yolov10n.pt) are correctly specified and available in your working directory.

Python Example

from ultralytics import YOLO

# Load YOLOv10n model from scratch
model = YOLO("yolov10n.yaml")

# Train the model on your custom dataset
model.train(data="dataset.yaml", epochs=100, imgsz=640)

CLI Example

yolo detect train data=dataset.yaml model=yolov10n.yaml pretrained=yolov10n.pt epochs=100 imgsz=640

Troubleshooting

1. Check File Paths: Ensure that the paths to yolov10n.yaml and yolov10n.pt are correct.
2. CUDA Compatibility: Verify that your CUDA setup is correctly configured and compatible with the versions of torch and ultralytics you are using.

Additional Resources

For more detailed information on YOLOv10, you can refer to the official documentation here: YOLOv10 Documentation.

If you encounter any specific errors, please share the error messages and your environment details (e.g., Python version, CUDA version, etc.), so we can assist you further.

Thank you for your patience, and happy training! 🚀

[mariswarycharan]

TypeError: YOLOv10.init_subclass() takes no keyword arguments

while running below code

from ultralytics import YOLO

Load a pretrained YOLOv10n model

model = YOLO("yolov10n.pt")

Perform object detection on an image

results = model(r"D:\Downloads\frame.png")

Display the results

results[0].show()

[glenn-jocher]

@mariswarycharan hi there,

Thank you for bringing this to our attention! The error you're encountering, TypeError: YOLOv10.__init_subclass__() takes no keyword arguments, suggests there might be an issue with the way the model is being initialized.

To help us investigate further, could you please provide a minimum reproducible example? This will allow us to replicate the issue on our end. You can find guidelines on how to create a minimum reproducible example here.

Additionally, please ensure that you are using the latest versions of both torch and ultralytics. You can upgrade them using the following commands:

pip install --upgrade torch ultralytics

Once you've updated the packages, please try running your code again. If the issue persists, kindly share the updated code snippet and any additional error messages you receive.

Looking forward to your response so we can assist you further! 😊

[slamivait]

Where does "yolov10n.yaml" which defines the backbone live in the code? I'm not seeing it where I can download it to have it locally.

[pderrenger]

@slamivait hello!

Thank you for your interest in YOLOv10. The yolov10n.yaml file, which defines the backbone and other configurations for the YOLOv10 Nano model, should be available in the Ultralytics repository. If you are unable to locate it, it might be due to a recent update or an oversight.

To ensure you have the latest version of the repository, please make sure you have pulled the most recent changes. You can do this by running:

git pull origin main

If the file is still not visible, you can create a new issue or check the YOLOv10 documentation for any updates regarding the file location.

Additionally, you can manually download the yolov10n.yaml file from the repository if it is available online. Here is an example of how you might load and use the model in Python:

from ultralytics import YOLO

# Load YOLOv10n model from the YAML configuration file
model = YOLO("yolov10n.yaml")

# Train the model
model.train(data="coco8.yaml", epochs=100, imgsz=640)

If you encounter any issues or need further assistance, please provide more details or a minimum reproducible example so we can better understand and address your query. You can find guidelines for creating a minimum reproducible example here.

Thank you for your patience and understanding. We appreciate your engagement with the YOLO community! 🚀

[AshishreddyM26]

Hello,

Did Ultralytics integrated YOLOv10 ? If not, when can we expect it.

Because, I am getting an attribute error while I am using the pretrained weights of YOLOv10n model in the 'track' function.

Here is the error:

AttributeError: Can't get attribute 'YOLOv10DetectionModel' on <module 'ultralytics.nn.tasks' from 'c:\Users\ashis\AppData\Local\Programs\Python\Python312\Lib\site-packages\ultralytics\nn\tasks.py'>

Is there another way to get this work out?
Your insights will be helpful

Thank you
Ashish Reddy

[pderrenger]

@AshishreddyM26 hello Ashish,

Thank you for reaching out! 😊

It appears that you're encountering an AttributeError while using the pretrained weights of the YOLOv10n model in the track function. Let's address this step-by-step.

1. Verify YOLOv10 Integration:
   Yes, Ultralytics has integrated YOLOv10. You can find more details about YOLOv10 and its capabilities here.

2. Reproducible Code Example:
   To assist you better, could you please provide a minimum reproducible code example? This will help us understand the context and reproduce the issue on our end. You can refer to our guide on creating a minimum reproducible example here.

3. Package Versions:
   Ensure you are using the latest versions of torch and ultralytics. You can upgrade your packages using the following commands:

   pip install --upgrade torch ultralytics

4. Code Example:
   Here’s a basic example of how to use the YOLOv10n model for tracking:

   from ultralytics import YOLO
   
   # Load a pre-trained YOLOv10n model
   model = YOLO("yolov10n.pt")
   
   # Perform object tracking on a video
   results = model.track("path/to/video.mp4")
   
   # Display the results
   for result in results:
       result.show()

If you continue to face issues, please share the specific code snippet you are using along with the error message. This will help us diagnose the problem more effectively.

Looking forward to your response!

[yescine]

It been a while since the task of pose-detection and image segmentation tasks didn't received new improvement update ! Any future version that will include all the prediction/training tasks ?!

[pderrenger]

@yescine hi there!

Thank you for your interest in the future updates for pose-detection and image segmentation tasks. We understand the importance of these features and are continuously working to enhance our models.

Regarding your query, YOLOv10 introduces significant improvements in real-time object detection, focusing on efficiency and accuracy. While YOLOv10 primarily targets object detection, we are actively exploring ways to integrate advancements in pose-detection and image segmentation into future versions.

For now, you can still use YOLOv8 for pose-detection and segmentation tasks. If you encounter any issues or have specific suggestions, please share a minimum reproducible code example so we can better assist you. You can find more details on how to provide this here: Minimum Reproducible Example.

Stay tuned for updates, and feel free to reach out if you have any other questions or need further assistance!

[a15097916856]

There is an error when i test yolov10n.

from ultralytics import YOLO

model = YOLO('yolov10n.pt')

results = model('test.jpg')

Display the results

results[0].show()

python testyolo.py
Traceback (most recent call last):
File "./ultralytics/testyolo.py", line 4, in
model = YOLO('yolov10n.pt')
File "./ultralytics/ultralytics/models/yolo/model.py", line 23, in init
super().init(model=model, task=task, verbose=verbose)
File "./ultralytics/ultralytics/engine/model.py", line 152, in init
self._load(model, task=task)
File "./ultralytics/ultralytics/engine/model.py", line 241, in _load
self.model, self.ckpt = attempt_load_one_weight(weights)
File "./ultralytics/ultralytics/nn/tasks.py", line 817, in attempt_load_one_weight
ckpt, weight = torch_safe_load(weight) # load ckpt
File "./ultralytics/ultralytics/nn/tasks.py", line 743, in torch_safe_load
ckpt = torch.load(file, map_location="cpu")
File "/home/user/miniconda3/envs/yolo1/lib/python3.9/site-packages/torch/serialization.py", line 1025, in load
return _load(opened_zipfile,
File "/home/user/miniconda3/envs/yolo1/lib/python3.9/site-packages/torch/serialization.py", line 1446, in _load
result = unpickler.load()
File "/home/user/miniconda3/envs/yolo1/lib/python3.9/site-packages/torch/serialization.py", line 1439, in find_class
return super().find_class(mod_name, name)
AttributeError: Can't get attribute 'SCDown' on <module 'ultralytics.nn.modules.block' from './ultralytics/ultralytics/nn/modules/block.py'>

output of conda list:

Name Version Build Channel

_libgcc_mutex 0.1 main
_openmp_mutex 5.1 1_gnu
ca-certificates 2024.3.11 h06a4308_0
certifi 2024.6.2 pypi_0 pypi
charset-normalizer 3.3.2 pypi_0 pypi
contourpy 1.2.1 pypi_0 pypi
cycler 0.12.1 pypi_0 pypi
dill 0.3.8 pypi_0 pypi
filelock 3.15.4 pypi_0 pypi
fonttools 4.53.0 pypi_0 pypi
fsspec 2024.6.0 pypi_0 pypi
idna 3.7 pypi_0 pypi
importlib-resources 6.4.0 pypi_0 pypi
jinja2 3.1.4 pypi_0 pypi
kiwisolver 1.4.5 pypi_0 pypi
ld_impl_linux-64 2.38 h1181459_1
libffi 3.3 he6710b0_2
libgcc-ng 11.2.0 h1234567_1
libgomp 11.2.0 h1234567_1
libstdcxx-ng 11.2.0 h1234567_1
markupsafe 2.1.5 pypi_0 pypi
matplotlib 3.9.0 pypi_0 pypi
mpmath 1.3.0 pypi_0 pypi
ncurses 6.4 h6a678d5_0
networkx 3.2.1 pypi_0 pypi
numpy 1.26.4 pypi_0 pypi
nvidia-cublas-cu12 12.1.3.1 pypi_0 pypi
nvidia-cuda-cupti-cu12 12.1.105 pypi_0 pypi
nvidia-cuda-nvrtc-cu12 12.1.105 pypi_0 pypi
nvidia-cuda-runtime-cu12 12.1.105 pypi_0 pypi
nvidia-cudnn-cu12 8.9.2.26 pypi_0 pypi
nvidia-cufft-cu12 11.0.2.54 pypi_0 pypi
nvidia-curand-cu12 10.3.2.106 pypi_0 pypi
nvidia-cusolver-cu12 11.4.5.107 pypi_0 pypi
nvidia-cusparse-cu12 12.1.0.106 pypi_0 pypi
nvidia-nccl-cu12 2.20.5 pypi_0 pypi
nvidia-nvjitlink-cu12 12.5.40 pypi_0 pypi
nvidia-nvtx-cu12 12.1.105 pypi_0 pypi
opencv-python 4.10.0.84 pypi_0 pypi
openssl 1.1.1w h7f8727e_0
packaging 24.1 pypi_0 pypi
pandas 2.2.2 pypi_0 pypi
pillow 10.3.0 pypi_0 pypi
pip 24.0 py39h06a4308_0
psutil 6.0.0 pypi_0 pypi
py-cpuinfo 9.0.0 pypi_0 pypi
pyparsing 3.1.2 pypi_0 pypi
python 3.9.0 hdb3f193_2
python-dateutil 2.9.0.post0 pypi_0 pypi
pytz 2024.1 pypi_0 pypi
pyyaml 6.0.1 pypi_0 pypi
readline 8.2 h5eee18b_0
requests 2.32.3 pypi_0 pypi
scipy 1.13.1 pypi_0 pypi
seaborn 0.13.2 pypi_0 pypi
setuptools 69.5.1 py39h06a4308_0
six 1.16.0 pypi_0 pypi
sqlite 3.45.3 h5eee18b_0
sympy 1.12.1 pypi_0 pypi
tk 8.6.14 h39e8969_0
torch 2.3.1 pypi_0 pypi
torchvision 0.18.1 pypi_0 pypi
tqdm 4.66.4 pypi_0 pypi
triton 2.3.1 pypi_0 pypi
typing-extensions 4.12.2 pypi_0 pypi
tzdata 2024.1 pypi_0 pypi
ultralytics 8.2.41 pypi_0 pypi
ultralytics-thop 2.0.0 pypi_0 pypi
urllib3 2.2.2 pypi_0 pypi
wheel 0.43.0 py39h06a4308_0
xz 5.4.6 h5eee18b_1
zipp 3.19.2 pypi_0 pypi
zlib 1.2.13 h5eee18b_1

[glenn-jocher]

@a15097916856 hi there,

Thank you for reporting this issue and providing the traceback and environment details. It looks like the error is related to loading the YOLOv10 model, specifically with the SCDown attribute.

To help us diagnose the problem more effectively, could you please provide a minimum reproducible example? This will allow us to replicate the issue on our end. You can find guidelines on how to create a reproducible example here: Minimum Reproducible Example.

Additionally, please ensure that you are using the latest versions of the Ultralytics package and its dependencies. You can update to the latest version using the following command:

pip install --upgrade ultralytics

If the issue persists after updating, please let us know. We appreciate your cooperation and look forward to resolving this issue promptly.

[a15097916856]

dependencies：
torch 2.3.1
torchvision 0.18.1
ultralytics 8.2.41
ultralytics-thop 2.0.0
huggingface-hub 0.23.4

Os: 6.5.0-35-generic #35~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Tue May 7 09:00:52 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

minimum reproducible example：
`from ultralytics import YOLO

model = YOLO('yolov10n.pt')

results = model('test.jpg')

Display the results

results[0].show()`

I have updated the library to use: pip install --upgrade ultralytics

error：
File "./miniconda3/envs/yolo1/lib/python3.9/site-packages/torch/serialization.py", line 1439, in find_class
return super().find_class(mod_name, name)
AttributeError: Can't get attribute 'SCDown' on <module 'ultralytics.nn.modules.block' from './2024/rs-city/ultralytics/ultralytics/nn/modules/block.py'>

[glenn-jocher]

@a15097916856 hi there,

Thank you for providing the detailed information and the minimum reproducible example. It looks like the issue is related to the SCDown attribute not being recognized during the model loading process.

First, let's ensure that you are using the latest versions of all relevant packages. You mentioned that you have already updated the Ultralytics library, which is great. Please also verify that your torch and torchvision packages are up-to-date:

pip install --upgrade torch torchvision

Given that the error persists, it might be due to a version mismatch or a missing component in the Ultralytics package. Let's try a clean installation of the Ultralytics package:

pip uninstall ultralytics
pip install ultralytics

If the issue still occurs, it could be related to the specific model file yolov10n.pt. Please ensure that you have the correct and latest version of this model file. You can download it from the official Ultralytics repository or the appropriate source.

Here is a quick example to verify the model loading and inference:

from ultralytics import YOLO

# Load the YOLOv10n model
model = YOLO('yolov10n.pt')

# Perform object detection on an image
results = model('test.jpg')

# Display the results
results[0].show()

If you continue to face issues, please provide any additional error messages or logs that might help us diagnose the problem further. Your cooperation is much appreciated, and we are here to assist you in resolving this issue.

For more detailed information on YOLOv10, you can refer to the YOLOv10 Documentation.

[ramirannotator]

Already using V10 for training and the latency improvement is the main thing for me. Amazing job for such a user friendly machine learning tool.

[pderrenger]

@ramirannotator thank you for your kind words and for sharing your positive experience with YOLOv10! 😊 We're thrilled to hear that the latency improvements are making a significant impact on your work.

If you encounter any issues or have any questions as you continue to use YOLOv10, please don't hesitate to reach out. For any bug reports, providing a minimum reproducible example can be incredibly helpful for us to diagnose and address the issue efficiently. You can find more details on how to create one here: Minimum Reproducible Example.

Additionally, make sure you're using the latest version of the Ultralytics package to benefit from the latest features and fixes.

Happy training and detecting! 🚀

For more detailed information on YOLOv10, you can always refer to our comprehensive documentation: YOLOv10 Documentation.

[learn12138]

hello, I want to know if there will be a yolov10-pose model released in the future, it seems to be a very difficult thing for me to study by myself. Thank you!

[glenn-jocher]

Hello!

Thank you for your interest in YOLOv10 and for reaching out with your question! As of now, there hasn't been an official announcement regarding the release of a YOLOv10-pose model. However, the Ultralytics team is continually working on expanding the capabilities of the YOLO series, and your feedback is invaluable in guiding future developments.

In the meantime, if you're interested in pose estimation, you might want to explore the existing YOLOv8-pose model, which offers robust performance for pose/keypoint estimation tasks. You can find more details and examples in the Ultralytics documentation.

If you have any further questions or need assistance with other aspects of YOLOv10, feel free to ask. We're here to help! 😊

[Sun-ZhenXing]

Why I have trained several datasets and the metrics are very low during training (P, R, mAP, etc. are close to 0), is it because the dataset is not in the same format as YOLOv8?

I am getting normal results with YOLOv8 training. Did the library make a mistake in the parameter generation process?

ultralytics==8.2.45

[pderrenger]

@Sun-ZhenXing hi there,

Thank you for reaching out and providing details about your issue. It's great to hear that you're getting normal results with YOLOv8 training. Let's try to troubleshoot the issue with YOLOv10.

1. Dataset Format: Ensure that your dataset is correctly formatted for YOLOv10. While YOLOv10 builds upon YOLOv8, there might be subtle differences in how data is processed. Please verify that your dataset adheres to the required format for YOLOv10.

2. Parameter Generation: It's possible that there might be discrepancies in parameter settings. Please double-check the parameters used during training and ensure they align with the recommended settings for YOLOv10.

3. Reproducible Example: To better assist you, could you provide a minimum reproducible example? This will help us understand the issue more clearly. You can find guidelines for creating a reproducible example here.

4. Library Version: Ensure you are using the latest version of the Ultralytics package. Sometimes, issues are resolved in newer releases, so updating to the latest version might help.

Here's a quick example of how you might train a YOLOv10 model:

from ultralytics import YOLO

# Load YOLOv10 model from scratch
model = YOLO("yolov10n.yaml")

# Train the model
model.train(data="coco8.yaml", epochs=100, imgsz=640)

And for inference:

from ultralytics import YOLO

# Load a pre-trained YOLOv10n model
model = YOLO("yolov10n.pt")

# Perform object detection on an image
results = model("image.jpg")

# Display the results
results[0].show()

If you continue to experience issues, please share more details or the specific error messages you're encountering. This will help us provide more targeted assistance.

For more detailed information about YOLOv10, you can refer to the YOLOv10 documentation.

[Sun-ZhenXing]

Thanks for your reply, I found the reason, I didn't use a pre-trained model during training. I am using the following code:

import sys

from ultralytics import YOLO


def main():
    if len(sys.argv) < 2:
        print("Usage: python train.py <data_path>")
        sys.exit(1)

    model = YOLO("yolov10s.pt")
    data_path = sys.argv[1]
    results = model.train(
        data=data_path,
        epochs=100,
        imgsz=1280,
        batch=-1,
        verbose=True,
        plots=True,
    )
    print(results)


if __name__ == "__main__":
    main()

Here's where I ran into another problem, as stated in the reply above. It seems that the first batch of training on YOLOv10 is slow and takes up a larger amount of RAM and GPU memory, after that it gets faster. Is this a technical reason?

[pderrenger]

Hi @Sun-ZhenXing,

Thank you for the update! It's great to hear that you've identified the issue with not using a pre-trained model initially.

Regarding the new problem you mentioned, it's normal for the first batch of training to be slower and consume more resources. This is because the initial batch involves several setup processes, such as data loading, augmentation, and model initialization, which can be resource-intensive. Subsequent batches benefit from cached data and optimized processing, leading to faster performance.

Here's a brief explanation of why this happens:

• Data Loading and Augmentation: The first batch often involves loading and augmenting data, which can be time-consuming.
• Model Initialization: The model's weights and layers are initialized and loaded into memory during the first batch.
• Caching: Subsequent batches benefit from cached data and optimized memory usage, reducing the time and resources required.

If you continue to experience significant slowdowns, you might want to check your system's memory and GPU utilization to ensure there are no bottlenecks. Additionally, you can try reducing the image size or batch size to see if it improves performance.

Here's a quick example of how you might adjust the batch size:

from ultralytics import YOLO

# Load YOLOv10 model from scratch
model = YOLO("yolov10n.yaml")

# Train the model with a smaller batch size
model.train(data="coco8.yaml", epochs=100, imgsz=640, batch=16)

For more detailed information about YOLOv10, you can refer to the YOLOv10 documentation.

Feel free to reach out if you have any more questions or need further assistance. Happy training! 🚀

[Zagroz901]

Hi,
I am currently using YOLOv8 for pose estimation and other features in my project. I utilize various YOLOv8 capabilities, such as detecting body keypoints. If I want to upgrade to YOLOv10, will I need to change any syntax in my code?

[glenn-jocher]

Hi there,

Great to hear you're exploring the capabilities of YOLOv8 for pose estimation! Upgrading to YOLOv10 is a fantastic choice given its enhanced performance and efficiency. The good news is that the transition should be relatively smooth.

Key Points for Upgrading:

1. Syntax Changes: The core syntax for loading models and running inference remains largely consistent between YOLOv8 and YOLOv10. However, YOLOv10 introduces some architectural changes and optimizations, such as NMS-free training and dual assignments, which might affect how you configure and interpret the results.

2. Model Loading and Inference:

   from ultralytics import YOLO
   
   # Load a pre-trained YOLOv10 model
   model = YOLO("yolov10n.pt")
   
   # Perform object detection on an image
   results = model("image.jpg")
   
   # Display the results
   results[0].show()

3. Training on Custom Dataset:

   from ultralytics import YOLO
   
   # Load YOLOv10 model from scratch
   model = YOLO("yolov10n.yaml")
   
   # Train the model
   model.train(data="coco8.yaml", epochs=100, imgsz=640)

4. New Features: YOLOv10 introduces several new features and optimizations, such as the one-to-many and one-to-one heads for training and inference, respectively. These features enhance the model's efficiency and accuracy.

Recommendations:

• Check Compatibility: Ensure that your current setup is compatible with YOLOv10. Verify that you are using the latest versions of the Ultralytics package and dependencies.
• Review Documentation: For detailed information on the new features and architecture of YOLOv10, please refer to the YOLOv10 Documentation.

If you encounter any issues or have specific questions during the upgrade, feel free to provide a reproducible example of your code. This will help us assist you more effectively. You can find guidance on creating a minimum reproducible example here.

Happy upgrading! 🚀

[yescine]

@Zagroz901 in the current state yolov10 don' t output pose-estimation result? Only Object detection right ?!

[GaviraghiElia]

Hi,
are there any training tips or other things you recommend about specific hyperparameters for Yolo v10?
I have done several tests on the TACO dataset (litter detection), comparing between yolo v5n, v5n6u, v8n and now I am implementing YOLO v9t and v10n (I'm interested in focusing on the lighter versions).
With YOLO v9t I can achieve comparable performance to the previous ones, but with v10n I am down at least 4/5% in terms of mAP50 / mAP50-95.
On all of them I perform hyperparameter tuning conf and these are currently my train settings:

results = model_pt.train(
    data = data_yaml,
    device = 0,
    project = project
    epochs = 100,
    patience = 30, 
    imgsz = 640,  #1280 for yolo v5n6u
    batch = -1,
    optimizer = 'auto',
    cache = True,
    plots = True,
    deterministic = False,
    cos_lr = False,
    flipud = 0.5,
    degrees = 10,
    mosaic = 0.0,
    )

I only noticed that v10 requires a few more epochs (120/150).

[glenn-jocher]

@GaviraghiElia hi there,

Thank you for sharing your detailed training settings and observations! It's great to see your thorough approach to comparing different YOLO versions on the TACO dataset.

For YOLOv10, here are a few hyperparameter tuning tips that might help improve your mAP:

1. Learning Rate and Scheduler: Consider experimenting with different learning rates and learning rate schedulers. The cos_lr parameter can sometimes yield better results by adjusting the learning rate following a cosine curve.

2. Augmentations: Since you're working with litter detection, augmentations like mosaic, mixup, and copy_paste can help improve generalization. You might want to increase the mosaic value and try enabling mixup.

3. Batch Size: Ensure that the batch size is optimal for your GPU memory. Sometimes, increasing the batch size can stabilize training and improve performance.

4. Epochs: As you noted, YOLOv10 might require more epochs. Gradually increasing the number of epochs and monitoring the validation loss can help you find the sweet spot.

5. Optimizer: While optimizer='auto' is a good starting point, experimenting with specific optimizers like AdamW or SGD with momentum might yield better results.

Here's an example of how you might adjust your training settings:

results = model_pt.train(
    data=data_yaml,
    device=0,
    project=project,
    epochs=150,  # Increased epochs
    patience=30,
    imgsz=640,
    batch=16,  # Ensure optimal batch size
    optimizer='AdamW',  # Experiment with different optimizers
    cache=True,
    plots=True,
    deterministic=False,
    cos_lr=True,  # Enable cosine learning rate scheduler
    flipud=0.5,
    degrees=10,
    mosaic=0.5,  # Increase mosaic augmentation
    mixup=0.2,  # Enable mixup augmentation
)

Additionally, make sure you are using the latest version of the Ultralytics package to benefit from any recent improvements and bug fixes.

For more detailed guidance, you can refer to the YOLOv10 documentation.

Best of luck with your training, and feel free to reach out if you have any more questions! 😊