Decouple semantic segmentation models

[kprokofi]

Summary

Eliminate mmseg dependency for models, decouple all models, introduce updated model structure, added support for torchvision data pipeline

Unit tests will be added with next PRs.

How to test

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have added e2e tests for validation.
• 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

[kprokofi]

Strange error on CI with CUDA

Locally, all unit tests are passed with tox

[jaegukhyun]

Overall LGTM.
Could you share some regression tests between decoupling?
In some case torchvision data pipeline slower than mm's pipeline since mm use opencv and it is faster than torch tensor ops.
So we need regression tests results for accuracy and e2e time.

[jaegukhyun]

And it seems unit tests run on cpu machine. So I think you need to fix unit tests failure on cpu machine.

[codecov]

Codecov Report

Attention: Patch coverage is 68.66097% with 220 lines in your changes are missing coverage. Please review.

Project coverage is 76.59%. Comparing base (4c18ab1) to head (7807455).

Files	Patch %	Lines
src/otx/algo/segmentation/backbones/mscan.py	61.71%	67 Missing ⚠️
src/otx/algo/segmentation/heads/ham_head.py	49.42%	44 Missing ⚠️
src/otx/algo/segmentation/heads/base_segm_head.py	73.13%	18 Missing ⚠️
...mentation/losses/cross_entropy_loss_with_ignore.py	46.87%	17 Missing ⚠️
src/otx/algo/segmentation/modules/aggregators.py	82.35%	12 Missing ⚠️
src/otx/core/model/segmentation.py	62.50%	12 Missing ⚠️
src/otx/algo/segmentation/heads/fcn_head.py	80.70%	11 Missing ⚠️
src/otx/algo/segmentation/backbones/dinov2.py	28.57%	10 Missing ⚠️
src/otx/algo/segmentation/base_model.py	81.25%	9 Missing ⚠️
src/otx/algo/segmentation/litehrnet.py	80.00%	7 Missing ⚠️
... and 3 more

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #3376      +/-   ##
===========================================
- Coverage    77.30%   76.59%   -0.72%     
===========================================
  Files          250      255       +5     
  Lines        22828    23281     +453     
===========================================
+ Hits         17648    17831     +183     
- Misses        5180     5450     +270     

Flag	Coverage Δ
py310	76.58% <68.66%> (-0.45%)	⬇️
py311	76.47% <68.66%> (-0.82%)	⬇️

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.

[kprokofi]

Tests are passed. I propose to merge this PR, since today is FF.
Please, feel free to leave comments to address in the following PRs before code freeze.

I compared algorithms on KITTI dataset.

iter_time is the same, data_train_time is worse on 0.002 sec., due to difference in resizing and augmentations, we have changed accuracy values about 2-4 % DICE score on this dataset. We need to align augmentations with mm*