Open lusen2000 opened 1 month ago
Hi, I have read the Robust Encoder part of the paper and tried to reproduce it. I used the loss in the paper and the DF2K paired dataset which utilized Real-ESRGAN degradation. But it seems that the vae only output dark images after a few iters of training. Would you enlighten me with more details of this part?
Hello,I also find after denoise_encoder and decoder, the input turns out dark output. Do you solve the problem?
Hi, I have read the Robust Encoder part of the paper and tried to reproduce it. I used the loss in the paper and the DF2K paired dataset which utilized Real-ESRGAN degradation. But it seems that the vae only output dark images after a few iters of training. Would you enlighten me with more details of this part?