Replication of Characteristic Visual Motifs of Indian Rural Art Forms using a Generative Adversarial Network

Abstract

The usage of Machine Learning in analyzing and creating art has accelerated substantially over the last couple of years, being used to mass produce novel pieces on a scale that artists would never have imagined feasible previously. This paper will focus on the study of an important Computer Vision tool- Generative Adversarial Networks (GANs) and whether or not it can learn and replicate the visually distinctive, defining characteristics of different forms of visual art, specifically rural Indian art forms. Indian rural art is a rich, underexplored area of visual media that consists of highly stylized, distinctive characteristics. We trained GANs to learn, abstract, and reproduce the visual motifs of five different forms of Indian rural art to understand whether GANs can learn these defining artistic attributes unsupervised and further our understanding of the differences between human intuition and machine perception of art. The textural and feature extraction ability will be analyzed through human perception as well as Fréchet Inception Distance to evaluate the quality of the generated images. This will benefit artists by allowing them to create artworks in unexplored genres, as well as benefit commercial industries, such as home décor and fashion, through the creation of different products in various art forms in an efficient manner.