I try to train codeformer stage I , but D loss always to 0

[kelisiya]

I use V100 GPU and batch is 1 , the loss and network same as paper , and my code reference GPEN training code .

[suixin7777]

I use V100 GPU and batch is 1 , the loss and network same as paper , and my code reference GPEN training code .

i think may you can try some HQ image to train, or you can see the https://github.com/CompVis/taming-transformers