Add Self-SL configs for SegNext

[sungchul2]

Summary

This PR is just to add Self-SL configs for our new semantic segmentation model, SegNext.
Performance verification for these configs is required.
With this configs, Self-SL training normally works with the below script:

otx train otx/algorithms/segmentation/configs/ham_segnext_t/template.yaml \
    --train-data-roots=path/dataset \
    params \
    --algo_backend.train_type=Selfsupervised

Experimental results

• Dataset
  • pre-training : sampled whole images with person, car, bicycle classes
  • fine-tuning : sampled 8 images with person, car, bicycle classes

model	mode	Epoch	mDice
ham_segnext_t	supervised	90.20 ± 28.08	64.23 ± 14.25
ham_segnext_t	pre-trained	65.53 ± 25.82	64.92 ± 12.48 (+0.69)
ham_segnext_s	supervised	81.00 ± 39.99	60.28 ± 17.23
ham_segnext_s	pre-trained	74.55 ± 25.68	68.13 ± 9.98 (+7.85)
ham_segnext_b	supervised	89.60 ± 16.93	68.25 ± 6.77
ham_segnext_b	pre-trained	69.15 ± 26.28	68.87 ± 8.08 (+0.62)

How to test

$ tox -vv -e tests-seg-py310 -- tests/integration/cli/semantic_segmentation
================================================================= short test summary info ==================================================================
SKIPPED [1] tests/integration/cli/semantic_segmentation/test_segmentation.py:90: Segnext model doesn't support supcon training.
FAILED tests/integration/cli/semantic_segmentation/test_segmentation.py::TestSegmentationCLI::test_otx_hpo[Custom_Semantic_Segmentation_SegNext_s] - AssertionError
============================================= 1 failed, 45 passed, 1 skipped, 2 warnings in 1249.68s (0:20:49) =============================================

$ tox -vv -e tests-seg-py310 -- tests/e2e/cli/semantic_segmentation
================================================================= short test summary info ==================================================================
SKIPPED [3] tests/e2e/cli/semantic_segmentation/test_segmentation.py:241: Skip the majority of models to reduce PTQ running time.
SKIPPED [3] tests/e2e/cli/semantic_segmentation/test_segmentation.py:250: Skip the majority of models to reduce PTQ running time.
SKIPPED [3] tests/e2e/cli/semantic_segmentation/test_segmentation.py:259: Skip the majority of models to reduce PTQ running time.
================================================= 136 passed, 9 skipped, 2 warnings in 13479.97s (3:44:39) =================================================

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) 2023 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

[kprokofi]

Hi @sungchul2, do we have time to update this PR and check if it works correctly? We updated SegNext from experimental template.

[goodsong81]

Too nitpick, but could you please fix the license notice?

[sungchul2]

Too nitpick, but could you please fix the license notice?

@goodsong81 Updated all.

[sovrasov]

@sungchul2 thanks for the PR. Do you have any experimental results obtained with this configs?

[sungchul2]

@sungchul2 thanks for the PR. Do you have any experimental results obtained with this configs?

I updated one of many results. Some results showed dropped performance but they are usually ~-2.0 mDice.