search

cszn / DnCNN

Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising (TIP, 2017)
https://cszn.github.io/
1.47k stars 539 forks source link

Loss function in Keras code #71

Open KewJieLong opened 5 years ago

KewJieLong commented 5 years ago

Hi, in your paper, u stated that MSE is used to optimized the network. However in your code, it is using Sum Square Error. It that a mistake or there is specific reason for that? Mind to share some experience?

betty97 commented 5 years ago

Hi, I think the code should be K.sum(K.square(y_pred - y_true))/(2*128) in order to adopt the MSE loss of the paper (adding the 1/128 factor in the Sum Squared Error expression), but I also don't understand why they are implementing Sum Squared Error instead of MSE. Could anyone explain it? Thanks