wyf0912/SinSR

Question about reproduction.

Closed this issue · 2 comments

Does your inverse loss (in paper) decrease? I try to reproduce it, but it always be 1.0.

The inverse loss remains unchanged is abnormal.

Did you make any changes to the code and could you post the training log?

The inverse loss can decrease normally when I trained the models.

Besides, you can check the inverse loss of the provided checkpoints. I think the inverse loss in checkpoints shall be much smaller than 1.