Update SR div2k-dataset documents

[EdwardZhao1991]

Hi all,

Thanks for the awesome project.

I'm a beginner at super-resolution and I wand to train an ESDR model based on DIV2K dataset.

However, based on the document, I cannot find the val_set5 on the official dataset website. https://mmediting.readthedocs.io/en/latest/sr_datasets.html#div2k-dataset

1. Could you please tell me how to obtain these files?

2. Moreover, it would be great if you can provide a demo code to generate the annotations. https://github.com/open-mmlab/mmediting/blob/2ab95c484a1eed48fccdd8a899aa89fbfb6b312f/configs/restorers/edsr/edsr_x2c64b16_g1_300k_div2k.py#L86

3. Where is the code to compute the mean and std? https://github.com/open-mmlab/mmediting/blob/2ab95c484a1eed48fccdd8a899aa89fbfb6b312f/configs/restorers/edsr/edsr_x2c64b16_g1_300k_div2k.py#L15-L16

does it consider the zero regions? e.g., in medical images (CT/MR), many pixels are zero.

Thanks again for the great project.

[Yshuo-Li]

1. val_set5 is another dataset called Set5
2. You can crop images by https://github.com/open-mmlab/mmediting/blob/master/tools/data/super-resolution/div2k/preprocess_div2k_dataset.py
3. Mean and std are provided in EDSR. Besides, we cannot give firm advice on whether consider the zero regions in medical images.