Generating human faces by generative adversarial networks

Abstract

Over the years, computer vision improves significantly. From recognising and understanding what lies underneath an image, we can now generate images by modelling training distribution using generative adversarial network(GAN). Since then, researchers come out with various variants of GAN and ways to stabalize GAN training. This results in improved quality of generated image. The application of GAN has sparked the interest of many people. In this project, we first analyse the use of StarGAN, a unified generative adversarial network for multi-domain image-to-image translation task to generate human facial expressions. We also explore the possible use of StarGAN in cartoon character facial expression generation and video generation.