shangpapa
Results
2
comments of
shangpapa
AG loss uses weight ω. The use of weight ω serves two purposes, the fist of them is that, the degree of positive and negative sample imbalance is squared due...
I solved this problem by add detach() on version 1.8.0. ` recon_loss = F.l1_loss(fake_img1, real_img1.detach()) `