Paper-Implementations
Paper-Implementations copied to clipboard
errD_real 和 errD_fake 分别backward和 这两个loss加起来用errD去backward有区别吗
我试过按照代码中errD_fake和errD_real分别backward,D网络和G网络以及L1的loss都能下降,用errD去backward,errD_real迅速降到零,errD_fake一直在上升,这是为什么呢
Hi, have you already found the anwser? I have also met thie problem and i cant tell the difference when i use errD for the backpropogation since its only a sum so it shoulnt influence the result at all. Looking forward to your reply
@jianingwangind It is caused by label