Enable to use input_size at transforms in recipe

[eunwoosh]

Summary

This PR enables to use input_size at transforms in recipe.
It's for unified input size. So, if value of all init_args in recipe uses input_size, all necessary values are automatically updated by changing only input_size.
For example, let's assume that we have recipe as below.

  data:
    train_subset:
      transforms:
        - class_path: otx.core.data.transform_libs.torchvision.RandomResize
          init_args:
            scale
            - 1000
            - 1000
        - class_path: otx.core.data.transform_libs.torchvision.RandomCrop
          init_args:
            crop_size
            - 500
            - 500

It can be changed as below using input_size.

  data:
    input_size:
    - 1000
    - 1000
    train_subset:
      transforms:
        - class_path: otx.core.data.transform_libs.torchvision.RandomResize
          init_args:
            scale: $(inputs_size)
        - class_path: otx.core.data.transform_libs.torchvision.RandomCrop
          init_args:
            crop_size: $(input_size) * 0.5

As you can see above, $(input_size) should be used to notify to use input_size.
Only simple multiplication or division evaluation is supported. So, addition or subtraction can't be used.
Type of input_size can be int or tuple[int, int]. If int value is given, it's considered as tuple[int, int] with same value.
whether passing int or tuple[int, int] to transform class is decided based on type hint. If there is no type hint, tuple[int, int] is passed.

input_size can be overrided by subset config. For example, if we want to use (1000, 1000) input size at train and (800, 800) at validation, recipe below can be used.

  data:
    input_size:
    - 1000
    - 1000
    train_subset:
      transforms:
        - class_path: otx.core.data.transform_libs.torchvision.RandomResize
          init_args:
            scale: $(inputs_size)
    val_subset:
      input_size:
      - 800
      - 800
      transforms:
        - class_path: otx.core.data.transform_libs.torchvision.RandomResize
          init_args:
            scale: $(inputs_size)

Of course, setting input_size only in subset confg rather than data config is available.
But input_size can't use $(input_size). It's available only in transforms.

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

[goodsong81]

Quite nice feature for recipe users or developers.
Generally looks good, but it might be controversial if it's "too much".
Left minor comments from a code reader's viewpoint. Feel free to ignore.

[sungchul2]

Thanks for your hard work. :)
I left some comments, please take a look.
And I think this PR has to include updated recipes to check unit and integration tests that are affected by this change.
What do you think?