Adapt input size automatically based on dataset statistics

[goodsong81]

Summary

• Classification / Segmentation

  • Set input size based on typical large image size in dataset
  • If larger than the default size, keep default size
  • ex) There's no much impact as most of datasets have images > 224x224.

• Detection / Instance segmentation

  • Set input size based on typical large image size in dataset
  • Adjust input size based on typical small shape annotation size in dataset
  • If larger than the default size, keep default size
  • ex) "Peanut" dataset for object detection

• Set "Auto" as default for all models

How to test

[Example CMD]

otx train src/otx/algorithms/detection/configs/detection/cspdarknet_yolox/template.yaml --train-data-roots ./data/peanut --workspace outputs/load-resize-det/feat/load-resize-cache-detiseg/peanut_cspdarknet_yolox_256x256_cache2g params --learning_parameters.input_size Auto --algo_backend.mem_cache_size 1000000000

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

[sungmanc]

I left minor comments, BTW, this PR needs the documentation. Will it be done at the upcoming PR? And just wondering about the performance change between before (develop) and after (this patch), no big differences?

[eunwoosh]

Thanks for your work! It's a nice feature. I left some comments, please take a look.

[goodsong81]

I left minor comments, BTW, this PR needs the documentation. Will it be done at the upcoming PR? And just wondering about the performance change between before (develop) and after (this patch), no big differences?

Thank you for the feedback. I would add documentation in a dedicated PR.
In addition, I'm thinking of reverting the default mode of "Auto" but enable this option in integration test. Turning it will be also handled in other PR if decided.

[goodsong81]

I left minor comments, BTW, this PR needs the documentation. Will it be done at the upcoming PR? And just wondering about the performance change between before (develop) and after (this patch), no big differences?

Thank you for the feedback. I would add documentation in a dedicated PR. In addition, I'm thinking of reverting the default mode of "Auto" but enable this option in integration test. Turning it will be also handled in other PR if decided.

Ah, I've just updated the document.