search

XLearning-SCU / 2022-CVPR-AirNet

PyTorch implementation for All-In-One Image Restoration for Unknown Corruption (AirNet) (CVPR 2022)
162 stars 21 forks source link

I have a question about the dataset generating process. #11

Open penguin1109 opened 1 year ago

penguin1109 commented 1 year ago

Before the question, I really thank you for the great research on multi degradation restoration.

But while reading your paper, I was curious on two things.

  1. You generated the positive samples by cropping patches from the same image, and negative samples from another image. How can this dataset generation process make the CBDE to learn the different degradation rates on the same image?
  2. The contrastive learning is implemented with MoCo in your code, but would it be okay if I used just the basic contrastive learning method?

Thank you. :)

Cat-Guijun commented 9 months ago

HI, have you addressed the two problems you mentioned above? Thanks.