Text-to-Painting on a Large Variance Dataset with Sequential Generative Adversarial Networks

Abstract

Converting text descriptions to images using Generative Adversarial Networks has become a popular research area. Visually appealing images were generated in recent years successfully. We investigated the generation of artistic images on a custom-built large variance dataset, which includes training images with variations, for example, in shape, color, and content. These variations in images provide originality, which is an important factor for artistic essence. One major characteristic of our work is that we used keywords as image descriptions, instead of sentences. As a network architecture, we proposed a sequential Generative Adversarial Network model, which utilizes several techniques like Wasserstein loss, spectral normalization, and minibatch discrimination to have stable training curves. Ultimately, we were able to generate painting images, which have a variety of styles. We evaluated the quality of generated paintings by using Frechet Inception Distance score.