Update semantic segmentation transforms to use OTX's instead of torchvision

[sungchul2]

Summary

This PR includes:

• Update transforms to OTX's
• Check performance for reproducibility

  • before performance is from after in Recipe refinement (det, iseg, r-det, sseg, vpm, zsl) #3712
    before : 746d07e | after : 2e32819

    model		test/Dice	export/Dice
    dino_v2	before	0.90	0.90
    	after	0.90	0.90
    litehrnet_18	before	0.90	0.89
    	after	0.89	0.89
    litehrnet_s	before	0.89	0.89
    	after	0.88	0.88
    litehrnet_x	before	0.90	0.88
    	after	0.90	0.85
    segnext_b	before	0.91	0.89
    	after	0.91	0.90
    segnext_s	before	0.91	0.91
    	after	0.90	0.90
    segnext_t	before	0.90	0.89
    	after	0.89	0.88

**NOTE
When using RandomResize + RandomCrop + Pad (mmseg pipeline) instead of RandomResizedCrop (current, torchvision pipeline), it seems that sometimes very small GT region was extracted like below figure and it significantly affected training performance, especially segnext_t (test/Dice : 0.46 and export/Dice 0.45). Due to this issue, use RandomResizedCrop instead.

comp. image and GT	loss & dice curve

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

[eunwoosh]

Thanks for your work:) I left some comments. Please take a look.