Comparative Analysis of Generative Adversarial Networks and their Variants

Generative Adversarial Networks (GAN)[1] is a generative modeling approach with a potential to learn high dimensional, complex real data distribution. In particular, they don't depend on any assumptions about the conveyance and can produce real-like examples from inert space in a simple manner. This powerful property drives GAN[1] to be applied to different applications, for example, picture blend, picture quality altering, picture interpretation, space variation and other scholarly fields. While great outcomes have been approved by visual assessment, a few quantitative rules have developed as of late. In this paper, we aim to discuss the operations and objective functions of variants of GAN[1] but do not comprehend GAN[1] deeply or who wish to view GAN from various perspectives. In addition, we present the comparison of evaluation of the images generated from variants of GAN like DCGAN, FCC-GAN and more.