Zhu Jun-Yan Park Taesung Isola Phillip Efros Alexei A.

Abstract

Image-to-image translation is a class of vision and graphics problems wherethe goal is to learn the mapping between an input image and an output imageusing a training set of aligned image pairs. However, for many tasks, pairedtraining data will not be available. We present an approach for learning totranslate an image from a source domain $X$ to a target domain $Y$ in theabsence of paired examples. Our goal is to learn a mapping $G: X \rightarrow Y$such that the distribution of images from $G(X)$ is indistinguishable from thedistribution $Y$ using an adversarial loss. Because this mapping is highlyunder-constrained, we couple it with an inverse mapping $F: Y \rightarrow X$and introduce a cycle consistency loss to push $F(G(X)) \approx X$ (and viceversa). Qualitative results are presented on several tasks where pairedtraining data does not exist, including collection style transfer, objecttransfiguration, season transfer, photo enhancement, etc. Quantitativecomparisons against several prior methods demonstrate the superiority of ourapproach.