Add Semi-SL Instance Segmentation

[kprokofi]

Summary

This PR includes Semi-SL Instance Segmentation solution based on the Mean Teacher algorithm after conducted PoC .

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

[codecov]

Codecov Report

Patch coverage: 71.94% and project coverage change: +0.16% 🎉

Comparison is base (a6b809e) 81.18% compared to head (dc245d1) 81.35%.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #2444      +/-   ##
===========================================
+ Coverage    81.18%   81.35%   +0.16%     
===========================================
  Files          516      516              
  Lines        37610    37686      +76     
===========================================
+ Hits         30534    30658     +124     
+ Misses        7076     7028      -48     

Flag	Coverage Δ
py310	81.34% <71.94%> (+0.16%)	⬆️
py38	81.33% <71.94%> (+0.16%)	⬆️
py39	81.33% <71.94%> (+0.16%)	⬆️

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

Files Changed	Coverage Δ
...ion/adapters/mmdet/models/heads/custom_roi_head.py	36.06% <ø> (ø)
src/otx/core/data/adapter/base_dataset_adapter.py	90.05% <ø> (ø)
...ommon/adapters/mmcv/hooks/custom_model_ema_hook.py	75.94% <7.69%> (-13.45%)	⬇️
...ms/common/adapters/mmcv/hooks/mean_teacher_hook.py	35.29% <33.33%> (ø)
.../common/adapters/mmcv/hooks/dual_model_ema_hook.py	76.71% <36.36%> (-1.86%)	⬇️
...dapters/mmdet/models/heads/custom_fcn_mask_head.py	32.43% <50.00%> (+1.00%)	⬆️
...on/adapters/mmdet/models/detectors/mean_teacher.py	79.14% <79.14%> (ø)
...n/adapters/mmdet/models/losses/cross_focal_loss.py	89.55% <81.81%> (-3.79%)	⬇️
...rc/otx/algorithms/detection/adapters/mmdet/task.py	84.71% <100.00%> (ø)
src/otx/api/entities/shapes/polygon.py	97.18% <100.00%> (-2.82%)	⬇️

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

[kprokofi]

Below the pivot table with PoC and comparison with other candidates inside OTX

Method	Tiny Pascal	CityScapes 8% old annotations (20+ classes)**	Sea garbage segmentation	CityScapes 9% new annotations (8 classes	WGISD	TACO**	Mean metric
Baseline (supervised)	22.13 / 24.94	27.5/52.5	25.63 / 44.82	34.81 / 52.69	18.49 / 25.25	6.4 / 16.6	25.26 / 36.92
Enhanced Mean Teacher	24.84 / 27.93	33.23 / 56.51	36.78 / 51.32	40.08 / 57.12	40.01 / 45.26	8.5 / 19.21	35.43 / 45.41 (+10.17 / + 8.49)
MT + Noisy Boundaries Head	23.66 / 27.17		35.47 / 50.87	39.84 / 57.16	39.56 / 44.41		34.63 / 44.9
MT + IoU masks filtering	24.48 / 27.88		35.75 / 49.65	39.84 / 56.15	39.78 / 45.08		34.96 / 44.69
MT + Unbiased Teacher V2	24.32 / 26.59		35.30 / 49.38	39.67 / 56.82	39.36 / 44.88		34.66 / 44.42
MT + PseCo	22.91 / 25.1	30.43 / 53.56	32.89 / 48.38	38.72 / 55.97	38.38 / 42.58		33.23 / 43.01

[kprokofi]

There are mmdet metrics:

Method	Tiny Pascal	Sea garbage segmentation	CityScapes 9% new annotations (8 classes	WGISD	Mean metric
Baseline (supervised)	58.70 / 59.50	28.30 / 28.70	45.60 / 48.30	25.70 / 24.70	39.57 / 40.30
Enhanced Mean Teacher best	66.80 / 70.20	40.10 / 39.30	50.90 / 53.10	50.30 / 48.70	52.03 / 52.83 (+12.46 / +12.53)

[sungmanc]

Could you add a summary for this PR to help to understand the objective and contents of this PR before reviewing?

[kprokofi]

This PR includes Semi-SL Instance Segmentation solution based on the Mean Teacher algorithm after conducted PoC .

I added it in the description
Basically, this PR includes Semi-SL Instance Segmentation solution based on the Mean Teacher algorithm.

[sungmanc]

Thanks for the effort, I left few comments

[goodsong81]

"Unbiased Teacher" -> general "Mean Teacher" scheme looks good to me.
Hope this would benefit both of det & iseg.
I've left a few minor change requests & qeustions.
Thank you for the great work!