Open noobtoob4lyfe opened 1 year ago
The two models are slightly different in training, vanillagan uses GAN Type=gan and wgan has GAN Type=wgan. We now believe that in practice prototype_vanillagan can produce a slight blur to counteract erroneous textures without losing the overall sharpness, while prototype_wgan is more sharp in all frames. I'm told that prototype_wgan has some slight color variations in some scenes due to some mistakes I made when training wgan, but this doesn't seem to be a serious problem.
Hi I noticed that the 2 pretrained models you shared have contents that are the exact same sizes. Are they really different in some way?