Closed MartinMeliss closed 6 years ago
@MartinMeliss Color inversion is one common failure mode for unsupervised image translation. Sometimes, it goes away when train for more iterations but sometimes it does not. I recently added multi-scale discriminator support to the code. Hope this can reduce the chance that you see color inverted.
Probably worth mentioning that sometimes the images are obtained with inverted colors, the same behavior of the network I observed in the CycleGAN. Helps restart the training.