Self-Supervised Dense Consistency Regularization for Image-to-Image Translation

Unsupervised image-to-image translation has gained considerable attention due to the recent impressive progress based on generative adversarial networks (GANs). In this paper, we present a simple but effective regularization technique for improving GAN-based image-to-image translation. To generate images with realistic local semantics and structures, we suggest to use an auxiliary self-supervised loss, enforcing point-wise consistency of the overlapped region between a pair of patches cropped from a single real image during training discriminators of GAN. Our experiment shows that the dense consistency regularization improves performance substantially on various image-to-image translation scenarios. It also achieves extra performance gains by using jointly with recent instance-level regularization methods. Furthermore, we verify that the proposed model captures domain-specific characteristics more effectively with only small fraction of training data.