house-leo / realsr-ccl Goto Github PK

Thank you for this great work and thank you for sharing the code.

Could you please share the training code of this work? Thanks.

Is your model supervised training? How is your L1/perception loss calculated in your model?

• Paired images are needed to calculate L1, and the images in the dataset of NTIRE2020 Challenge are all unpaired ones. How do you calculate L1 loss? Does your network have a domain adaptation network? These contents are not involved clearly in your paper. How do you explain this? Are you using validation sets for paired images?
• In the same way, the perception loss also requires paired images, for the same reason as L1 above, which is not detailed here.