Improve polygon utils

[wonjuleee]

Summary

Creating dm.Polygon is not compute-efficient way because of running annotation validation.
So, through this PR, I would like to reuse the current dm.Polygon annotations instead.

With RTMDet-Tiny model on large instance segmentation dataset (5 times repeat), I can reduce 2.5% iteration time on average.

	epoch	e2e	iter time	f1
develop	30.2	326.5254	0.32453617	0.44547
fix_polygon	29	310.143	0.31644868	0.451077

<style> </style>

	develop	fix_polygon	reduction (%)
otx.core.data.transform_libs.torchvision.CachedMosaic:	0.28111	0.27406	2.507683577
otx.core.data.transform_libs.torchvision.RandomResize:	0.02781	0.02545	8.486345826
otx.core.data.transform_libs.torchvision.RandomCrop:	0.00561	0.00499	11.1882294
otx.core.data.transform_libs.torchvision.YOLOXHSVRandomAug:	0.00529	0.00513	3.088670277
otx.core.data.transform_libs.torchvision.RandomFlip:	0.00058	0.00044	24.97037152
otx.core.data.transform_libs.torchvision.Pad:	0.00109	0.00107	2.479535156
otx.core.data.transform_libs.torchvision.CachedMixUp:	0.01892	0.01944	-2.76476453
otx.core.data.transform_libs.torchvision.FilterAnnotations:	0.00186	0.00186	0.312082033
torchvision.transforms.v2._misc.ToDtype:	0.00122	0.00139	-13.44773689
torchvision.transforms.v2._misc.Normalize:	0.00256	0.00265	-3.36598766
SUM	0.346057	0.336452	2.775750417

How to test

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have ran e2e tests and there is no issues.
• I have added the description of my changes into CHANGELOG in my target branch (e.g., CHANGELOG in develop).​
• I have updated the documentation in my target branch accordingly (e.g., documentation in develop).
• I have linked related issues.

License

• I submit my code changes under the same Apache License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below).

# Copyright (C) 2024 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

[harimkang]

LGTM, i left minor comment

[eunwoosh]

Thanks for your work. Could you update integration test result? and I left some comments. Please take a look.

[sungchul2]

LGTM. But could you compare data time only? It doesn't look like the performance improvement is evident.
(Of course, I think there is good improvement because reallocating dm.Polygon which was a bottleneck disappeared.)

[codecov]

Codecov Report

Attention: Patch coverage is 88.23529% with 2 lines in your changes missing coverage. Please review.

Project coverage is 80.47%. Comparing base (3a44e46) to head (df46747).

Files	Patch %	Lines
src/otx/core/data/transform_libs/utils.py	88.23%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #3707      +/-   ##
===========================================
- Coverage    80.54%   80.47%   -0.08%     
===========================================
  Files          255      255              
  Lines        25835    25829       -6     
===========================================
- Hits         20810    20787      -23     
- Misses        5025     5042      +17     

Flag	Coverage Δ
py310	80.47% <88.23%> (-0.06%)	⬇️
py311	?

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wonjuleee]

Thanks for your work. Could you update integration test result? and I left some comments. Please take a look.

I have just run instance_segmentation related tests here.

[kprokofi]

@wonjuleee , could you briefly describe what exactly here helps to reduce time and raise metrics? Because as I see it is just some refactoring done. Did you check it with different seeds and data?
UPD: I got it, thanks :)

[wonjuleee]

@wonjuleee , could you briefly describe what exactly here helps to reduce time and raise metrics? Because as I see it is just some refactoring done. Did you check it with different seeds and data? UPD: I got it, thanks :)

Hi Kirill, thank you for the question. Previously, it used to create a new dm.Polygon annotation for each annotation, but it takes some converter logics described in https://github.com/openvinotoolkit/datumaro/blob/6a9227694fc4bac0883269974118dfd318bc573d/src/datumaro/components/annotation.py#L687. So, I turned to reuse the existing dm.Polygon annotation and update only the points for the augmentations (random crop/resize/flip).