Applying Deep Generative Models to Portrait Art Generation: A Comparative Study of GANs and VAEs

Abstract

Deep Generative Models (DGMs) have emerged as powerful tools for generating diverse and realistic data across various domains. This paper presents a comprehensive systematic review of existing DGMs, discovering their methodologies, architectures, and applications. We delve into the fundamental concepts of Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), highlighting their respective strengths and weaknesses. Furthermore, we provide a detailed case study focusing on the utilization of GANs and VAEs for generating images of portrait art. By employing a dataset of portrait artworks, we demonstrate the capabilities of these DGMs in capturing the latent representation to generate new art’s. Through a comparative analysis of the generated results, we evaluate the likelihood and the inception score achieved by each model. By diving into theoretical insights with practical experimentation, this paper offers valuable insights into DGMs and their potential applications. The findings and discussions presented contribute to a deeper understanding of deep generative modeling techniques and pave the way for future advancements the field