a question about train - Githubissues

The-Learning-And-Vision-Atelier-LAVA / DASR

[CVPR 2021] Unsupervised Degradation Representation Learning for Blind Super-Resolution

MIT License

388 stars 50 forks source link

a question about train #11

Open Salmashi opened 3 years ago

Salmashi commented 3 years ago

The resolution of images in DF2K is relatively high. May I ask if you need to crop to a certain resolution before training?

LongguangWang commented 3 years ago

Hi @Salmashi, thanks for your interest in our work. We do not need to crop the HR images before training. Patches will be cropped online in the dataloader (https://github.com/LongguangWang/DASR/blob/main/data/multiscalesrdata.py#L157)

jiahong-fu commented 3 years ago

Hi @LongguangWang ,why did you extract two patches from each image?

LongguangWang commented 3 years ago

Hi @jiahong-fu, we extract two patches from one image to construct positive pairs (with the same degradation) for contrastive learning, as illustrated in Fig. 1.