From the group that brought you pix2pix, comes CycleGAN. CycleGAN learns to translate partial images from one domain or style to another (e.g. turning a horse into a zebra) but without requiring matching image pairs, a limitation of pix2pix. Removing the image pairs and relying on the network to learn a latent representation of each style makes this useful in the real world. If you’re interested in this area, CVAE-GAN is another similar paper from last week.
Horses to Zebras and Back Again With CycleGAN
Post · Mar 31, 2017 15:25 · Share on Twitter
Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain `X` to a target domain `Y` in the absence of paired examples. Our goal is to learn a mapping `G: X => Y` such that the distribution of images from `G(X)` is indistinguishable from the distribution $Y$ using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping `F: Y => X` and introduce a cycle consistency loss to push `F(G(X)) ~= X` (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.