Skip to content

Latest commit

 

History

History
22 lines (19 loc) · 1.69 KB

README.md

File metadata and controls

22 lines (19 loc) · 1.69 KB
Raw

How To Prepare Rank Dataset

Prepare perceptual data

  1. Prepare Three levels SR Models. You can download the [SRResNet (SRResNet_bicx4_in3nf64nb16.pth), SRGAN (SRGAN.pth), ESRGAN (ESRGAN_SuperSR.pth)] from Google Drive or Baidu Drive. You could place them in ./experiments/pretrained_models/.

  2. Download DIV2K and Flickr2K from Google Drive or Baidu Drive

  3. Generate Three level images using 'How to test' with codes/options/test/test_RankSRGAN.yml

Generate rank dataset

  1. Training dataset: Use ./datasets/generate_rankdataset/generate_rankdataset.m to generate three level training patchs.
  2. Validation dataset: Use ./datasets/generate_rankdataset/move_valid.py to generate three level patchs.
  3. Rank label: Use ./datasets/generate_rankdataset/generate_train_ranklabel.m to generate Training Rank label (NIQE). Use ./datasets/generate_rankdataset/generate_valid_ranklabel.m to generate Validation Rank label (NIQE).