Support specifying scales in preprocessing div2k dataset

[wileewang]

Motivation

We want build a data set like DIV2KRK which contain low-quality(LQ) images, ground-truth(GT) images and blur kernels and use it for training models like Deep Alternating Network(DAN), which means we also need to crop them into sub images. This PR create a scripts to generate this data set.

Modification

We create a new file preprocess_div2k_dataset_bsr.py, a copy of preprocess_div2k_dataset.py, and then add a function to generate blur kernel according to configuration, and we no longer use DIV2K LR images to crop into sub images but apply pipeline used in DAN's paper: GT-->blur-->down-sampling-->LQ, and we also stored corresponding blur kernel in .mat format using same base name.

[ckkelvinchan]

Hello, thank you for your contribution. May I ask why do you pre-generate the LR images instead of generating the LR during training? I think the degradations could be more diverse if we apply the degradations during training.

[wileewang]

Hello, thank you for your contribution. May I ask why do you pre-generate the LR images instead of generating the LR during training? I think the degradations could be more diverse if we apply the degradations during training.

Yes, you're right. Maybe I think pre-generating low-quality images is convenient and intuitive, and if we generate lq images during training, we will no longer need LR sub images generated by preprocess_div2k_dataset.py.

Thanks for your advice! I have a new idea. We add a new argument in preprocess_div2k_dataset.py to control whether generating LR sub images or not as more and more researches are performing more complex modeling of the degradation process, not just down-sampling. And we can degrade images during training.

[wileewang]

I have modified my pull request by

1. Removing preprocess_div2k_dataset_bsr.py
2. Adding a new argument --custom-degradation into preprocess_div2k_dataset.py to control whether crop bicubic LR images
3. Adding some illustrations into README.md and README_zh-CN.md

[ckkelvinchan]

I have modified my pull request by

1. Removing preprocess_div2k_dataset_bsr.py
2. Adding a new argument --custom-degradation into preprocess_div2k_dataset.py to control whether crop bicubic LR images
3. Adding some illustrations into README.md and README_zh-CN.md

I think you can simply use an argument scale to specify whether you want to process the downsampled images.

If scale = [], then only the HR images will be processed.

[wileewang]

I have modified my pull request by

1. Removing preprocess_div2k_dataset_bsr.py
2. Adding a new argument --custom-degradation into preprocess_div2k_dataset.py to control whether crop bicubic LR images
3. Adding some illustrations into README.md and README_zh-CN.md

I think you can simply use an argument scale to specify whether you want to process the downsampled images.

If scale = [], then only the HR images will be processed.

Thanks for your advice. I have updated it.

[codecov]

Codecov Report

Merging #472 (7013f70) into master (f9c9a95) will decrease coverage by 0.01%.
The diff coverage is n/a.

❗ Current head 7013f70 differs from pull request most recent head 8d40b78. Consider uploading reports for the commit 8d40b78 to get more accurate results

@@            Coverage Diff             @@
##           master     #472      +/-   ##
==========================================
- Coverage   80.56%   80.54%   -0.02%     
==========================================
  Files         190      190              
  Lines       10338    10338              
  Branches     1533     1533              
==========================================
- Hits         8329     8327       -2     
- Misses       1780     1781       +1     
- Partials      229      230       +1     

Flag	Coverage Δ
unittests	80.52% <ø> (-0.01%)	⬇️

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

Impacted Files	Coverage Δ
...ls/components/stylegan2/generator_discriminator.py	84.93% <0.00%> (-1.21%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f9c9a95...8d40b78. Read the comment docs.

[OpenMMLab-Coodinator]

Hi @wileewang ！First of all, we want to express our gratitude for your significant PR in this project. Your contribution is highly appreciated, and we are grateful for your efforts in helping improve this open-source project during your personal time. We believe that many developers will benefit from your PR.

We would also like to invite you to join our Special Interest Group (SIG) private channel on Discord, where you can share your experiences, ideas, and build connections with like-minded peers. To join the SIG channel, simply message moderator— OpenMMLab on Discord or briefly share your open-source contributions in the #introductions channel and we will assist you. Look forward to seeing you there! Join us ：https://discord.gg/raweFPmdzG

If you are Chinese or have WeChat，welcome to join our community on WeChat. You can add our assistant ：openmmlabwx. Please add "mmsig + Github ID" as a remark when adding friends：）
Thank you again for your contribution❤