Discriminator optimal transport

Akinori Tanaka

Abstract: Within a broad class of generative adversarial networks, we show that the discriminator optimization process increases a lower bound of the mean discrepancy dual to transport cost function in terms of Wasserstein distance between the target distribution $p$ and the generator distribution $p_G$. It implies that the discriminator around equilibrium can approximate optimal transport (OT) from $p_G$ to $p$. Based on some experiments and a bit of OT theory, we propose a discriminator optimal transport (DOT) scheme to improve generated images. We show that it improves inception score and FID of samples generated by un-conditional GAN trained by CIFAR-10, STL-10 and a public pre-trained model of conditional GAN by ImageNet.

Code Link: https://github.com/AkinoriTanaka-phys/DOT

CMT Num: 3691