Not an issue: Generator as Autoencoder and Loss = L1 Loss + Adversarial Loss

[nikhilchh]

I am thinking about an implementation where generator is an auto-encoder. So that we can do something like pix2pix in progressive-growing fashion.