Hi. Your code helped me a lot, but I was wondering the same thing.
If you don’t freeze the discriminator weights, doesn’t the batch on adversarial modify discriminator weights so that the it outputs 1 for generated image too? I don’t mean to freeze the weights once reached some metric, but i mean to freeze and unfreeze them at every epoch. Something like:
do batch on discriminator
freeze discriminator weights
do batch on adversarial
unfreeze discriminator weights
Is it right in both ways? Is there a better one?