Recent Advances of Generative Adversarial Networks

Abstract

Generative adversarial networks (GANs) have received great attention recently. GAN is an unsupervised learning method that learns representations without highly relying on annotations. GAN comprises two networks called generator and discriminator that training in a competitive process. GANs have made notable progress and promising performance in various applications including image inpainting, image super-resolution, and face synthesis. This paper aims to introduce an overview of GAN, including fundamentals, several up-to-date variants, applications, and challenges of GAN. Finally, we also provide readers with some solutions to mitigate issues existing in GANs.