About the loss function in source code

[Lan1eve]

Thanks for your great work!!!
I'm confused about one thing. The loss function in the open source code, in basic_restoer, only uses charbonnier_loss, and does not use the gradient loss mentioned in the paper. Did you use gradient loss for training? and if so, where is the corresponding code?

[Zhanghao Sun]

Hi, thanks for your interest in our paper! We found that using an additional gradient loss just increase the performance by a few percents, so we did not include the gradient loss in the repo.

Note that you should be able to mostly reproduce our training results just with charbonnier_loss (i.e., the training configs we provided in the repo). Please let us know if it does not work out.

[Lan1eve]

Thanks for your reply!