A Survey of Modern Deep Learning based Generative Adversarial Networks (GANs)

Abstract

GANs (Generative Adversarial Networks) are a type of deep learning generative model that has lately gained popularity in recent years. GANs can learn patterns from high-dimensional complex data, making them useful for image, audio and video processing. Nonetheless, there are several significant obstacles in the training of GANs, such as instability, mode collapse and non-convergence. To address these issues, researchers have developed a variety of GAN variations by rethinking network topology, modifying the form of goal functions, and changing optimization to precise methods in recent years. This paper describes a thorough analysis of the progress of GAN architecture and optimization solutions to improve its efficiency in various computer vision applications and challenges that are to be faced while implementing the model towards CV (computer vision) is described. It is believed that GAN is strong model and further researches are needed to work in this area to solve a variety of computer vision real time applications.