search

junyanz / CycleGAN

Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
Other
12.22k stars 1.94k forks source link

About the loss of the generator #152

Closed hello-trouble closed 2 years ago

hello-trouble commented 2 years ago

Dear professor: Thank you for your excellent paper and the project. I am puzzled about the unsupervised LSGAN loss in my own project .Both the loss of G and D are very little, almost near the zero. the figure of the losses are as follows: image image image The learning rate are set to 5e-5, Usually, the loss_D tend to be zero with a strong discriminator. However,In my project the loss_G_un is also very lillte, In my opinion, it may be caused by the weiht (0.5 is used) or the training data ? I am now trying a larger weight for the generator. Would you mind give me some advice ? Looking forward to your help. Have a good weekend

hello-trouble commented 2 years ago

Dear professor: So sorry to bother you. I have solved this problem yesterday. Thank you again for your excellent paper and projects.