Generative Adversarial Networks in Computer Vision

Generative adversarial networks (GANs) have been extensively studied in the past few years. Arguably their most significant impact has area of computer vision where great advances made challenges such as plausible image generation, image-to-image translation, facial attribute manipulation, and similar domains. Despite successes achieved to date, applying GANs real-world problems still poses challenges, three which we focus on here. These are follows: (1) generation high quality images, (2) diversity (3) stabilizing training. Focusing degree popular GAN technologies progress against these provide a detailed review state-of-the-art GAN-related research published scientific literature. We further structure this through convenient taxonomy adopted based variations architectures loss functions. While several reviews for presented none considered status field toward addressing practical relevant vision. Accordingly, critically discuss architecture-variant, loss-variant GANs, tackling challenges. Our objective is an overview well critical analysis terms application requirements. As do also compelling applications demonstrated considerable success along with some suggestions future directions. Codes related GAN-variants work summarized https://github.com/sheqi/GAN_Review.