KAIR
KAIR copied to clipboard
DnCNN training options
Hi, thanks for this great work! I have some questions regarding DnCNN model training. In the paper, the loss function applied is MSE Loss, may I know why it is stated that L1 loss is preferred in the training JSON?
Besides, I am well aware that there are scripts for the jpeg compression dataset, but how should I do if I want to create a custom dataset for Gaussian noise and jpeg compression?
Thank you very much.