@junyanz Sir, Is there any trick or better way to train a CycleGAN if I have very fewer data from one domain but sufficient data from the other domain?
One idea is to (1) train an auto-encoder for the domain with sufficient data. (2) share the generator across two domains. See some recent work for more details [1], [2].
@junyanz Sir, Is there any trick or better way to train a CycleGAN if I have very fewer data from one domain but sufficient data from the other domain?