About CycleGan.py

Open ZengZihang19 opened this issue 3 years ago • 1 comments

When running CycleGan.py, the values of dA_loss_fake and B_loss_fake are [0.999XXX, 0.0], is this normal? Moreover, D loss has been maintained at about 0.5, acc is 50%, and the adv of G loss has always been 0.

Sep 28 '22 11:09 ZengZihang19

It shall be fine. As a stable GAN will have a discriminator loss around 0.5. Besides, we've updated a pre-trained "generator.h5". You can directly use it for your experiment.

Oct 14 '22 04:10 Megum1