Use torchvision in MRCNN on CUDA

[sovrasov]

Summary

This could mitigate QAT instability on some datasets
Dataset: geti test data from the ticket
Model: ResNet50 MRCNN

Backend	iter time	total train time	f-score
mmcv	0.362	0:03:06	0.4021
torchvision	0.365	0:03:09	0.405

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

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 80.73%. Comparing base (76b89ea) to head (dee36bb).

Additional details and impacted files

@@                Coverage Diff                 @@
##           releases/1.6.0    #3347      +/-   ##
==================================================
- Coverage           80.76%   80.73%   -0.03%     
==================================================
  Files                 536      536              
  Lines               40476    40477       +1     
==================================================
- Hits                32689    32681       -8     
- Misses               7787     7796       +9     

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

[chuneuny-emily]

@sovrasov - Will it affect the speed of nncf optimization a lot? Should it only be applied to MaskRCNN-ResNet50 which was the offending thing? or all models?

[sovrasov]

@sovrasov - Will it affect the speed of nncf optimization a lot? Should it only be applied to MaskRCNN-ResNet50 which was the offending thing? or all models?

This PR is for testing purposes, I'll update the implementation later to restrict the scope of the changes. Also, I'm planning to benchmark performance and accuracy with this patch today

[sovrasov]

@chuneuny-emily please see the benchmark results mmcv vs torchvision in the PR description. torchvision is slightly slower, but I think this is acceptable.