Adversarial Self-Defense for Cycle-Consistent GANs

State-of-art methods learn the correspondence using large numbers of unpaired examples from both domains and are based on generative adversarial networks.