Add Mask, SuperResolution, Depth visualization features #747
Conversation
|
This should be merged after #746. |
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
I added Jupyter notebook examples for visualization in this PR. The related ticket is no. 94408. |
Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
function Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
vinnamkim
force-pushed
the
feature/truly-seg-visualizer
branch
from
978e383
to
e777304
Compare
Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
|
View / edit / reply to this conversation on ReviewNB wonjuleee commented on 2022-10-31T04:32:42Z Specifically, we are going to provide the example codes for instance segmentation and captioning tasks with MS-COCO 2017 dataset. vinnamkim commented on 2022-10-31T07:49:23Z Added. |
|
View / edit / reply to this conversation on ReviewNB wonjuleee commented on 2022-10-31T04:32:43Z Can we reuse the tests/assets instead of downloading the full dataset? vinnamkim commented on 2022-10-31T07:20:00Z I don't know whether there is a legal issue to upload some of images in COCO dataset to this repository. This downloading COCO dataset in notebook examples is same as yolov5/tutorial.ipynb at master · ultralytics/yolov5 (github.com). wonjuleee commented on 2022-10-31T08:09:49Z I don't think downloading all dataset for running this notebook is efficient. vinnamkim commented on 2022-10-31T08:19:30Z Out of efficiency issue, my opinion is that this notebook is intended to give users insight or teach them about how to use Datumaro with real-world datasets. However, I do not think that explaining this with artificially created small datasets or custom assets will not be of sufficient help to users. Especially because users are likely to make use of this notebook to solve their problem rather than just running it, it's good start point to use the real-world dataset as an example. vinnamkim commented on 2022-10-31T08:20:39Z In addition, it does not require many resources because COCO val dataset is <1G. wonjuleee commented on 2022-11-01T00:45:09Z If so, we need to download other datasets for other notebooks. My suggestion is making an option to download public datasets or prepare another separated notebook for downloading public datasets. Referring the notebook could be enough. On the other hand, Datumaro provides the downloading command through CLI. vinnamkim commented on 2022-11-01T00:53:48Z I guess download from CLI command only supports BBOX for now. https://github.com/openvinotoolkit/datumaro/blob/a8f3cc947594d7fe2f16a7f61a07a67150cfb78f/datumaro/components/extractor_tfds.py#L287-L302 Can we leave this to be a future work? (Current) download from source -> (Future) Download from Datumaro CLI wonjuleee commented on 2022-11-01T01:33:28Z Yes, sure. Let's take a look at CLI issues with another thread. |
|
I don't know whether there is a legal issue to upload some of images in COCO dataset to this repository. This downloading COCO dataset in notebook examples is same as yolov5/tutorial.ipynb at master · ultralytics/yolov5 (github.com). View entire conversation on ReviewNB |
Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com>
|
Added. View entire conversation on ReviewNB |
|
I don't think downloading all dataset for running this notebook is efficient. View entire conversation on ReviewNB |
|
Out of efficiency issue, my opinion is that this notebook is intended to give users insight or teach them about how to use Datumaro with real-world datasets. However, I do not think that explaining this with artificially created small datasets or custom assets will not be of sufficient help to users. Especially because users are likely to make use of this notebook to solve their problem rather than just running it, it's good start point to use the real-world dataset as an example. View entire conversation on ReviewNB |
|
In addition, it does not require many resources because COCO val dataset is <1G. View entire conversation on ReviewNB |
|
If so, we need to download other datasets for other notebooks. My suggestion is making an option to download public datasets or prepare another separated notebook for downloading public datasets. Referring the notebook could be enough. On the other hand, Datumaro provides the downloading command through CLI. View entire conversation on ReviewNB |
|
I guess download from CLI command only supports BBOX for now. https://github.com/openvinotoolkit/datumaro/blob/a8f3cc947594d7fe2f16a7f61a07a67150cfb78f/datumaro/components/extractor_tfds.py#L287-L302 Can we leave this to be a future work? (Current) download from source -> (Future) Download from Datumaro CLI View entire conversation on ReviewNB |
|
Yes, sure. Let's take a look at CLI issues with another thread. View entire conversation on ReviewNB |
I created a backlog ticket no. 95780 for it. |
…kit#747) * Add _draw_mask Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Add SR and Depth visualization Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Update changelog.md Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Fixing _draw_mask numpy slicing bug and revise test to wrap the original function Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Add Jupyter notebook example Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Add parameter descriptions to _draw() Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Add more explanation to notebook intro Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> * Remove draw_only_image flag Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> Signed-off-by: Kim, Vinnam <vinnam.kim@intel.com> Co-authored-by: Wonju Lee <wonju.lee@intel.com>
Summary
draw_only_image=True, down:draw_only_image=FalseHow to test
I added unit tests for changes.
Checklist
developbranchLicense
Feel free to contact the maintainers if that's a concern.