CycleGAN: cycle-consistency is the key
How CycleGAN learns to translate between two image domains without a single matched pair — and why the cycle-consistency loss, not the adversarial loss, is what makes it work.
~/blog/tags/
1 post
How CycleGAN learns to translate between two image domains without a single matched pair — and why the cycle-consistency loss, not the adversarial loss, is what makes it work.