We should add variable tracking across training sessions (when training a GAN multiple times). This is implemented in the autoencoder and we should port that to the GAN. One important feature is the actual epochs trained for each training session (e.g., if it crashes, there's no way to know how many epochs have been trained).
We should add variable tracking across training sessions (when training a GAN multiple times). This is implemented in the autoencoder and we should port that to the GAN. One important feature is the actual epochs trained for each training session (e.g., if it crashes, there's no way to know how many epochs have been trained).