A loss function of GAN to generate various images

Abstract

We propose a new loss function derived from variance of generated images of GAN(Generative Adversarial Networks), and applied the function to a generative model of GAN. Using the function, it is expected to avoid mode collapse and generate various images. In this study, GAN with the function was examined how various images can be generated and which parameter can contribute to the variance value through illustration dataset. Outcome of the study has potential ability relevant to auto-generators of illustration. Auto-generators of illustration can assist human creative activities and entertainment .