Giters

cszn / KAIR

Image Restoration Toolbox (PyTorch). Training and testing codes for DPIR, USRNet, DnCNN, FFDNet, SRMD, DPSR, BSRGAN, SwinIR

Home Page:https://cszn.github.io/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Dataset preperation for SWIN-IR

yossigandelsman opened this issue · comments

commented

Hi, I am trying to recreate the results from the paper.
I downloaded the two training datasets for "classical/lightweight image SR" from the provided links in your documentation.
Unfortunately, I only see 743 images in the DIV2K_train_HR directory under DIV2K and not 800. Did you also use the test set?
In the Flickr2K I only see the LR versions (Flickr2K_LR_bicubic, Flickr2K_LR_unknown). Should I download the HR images from somewhere else?

Thanks for the help!

commented

I think your file may be corrupted. I downloaded it just now and can find there are 900 images in DIV2K. You should use the first 800 images.