Face Generation and Editing with StyleGAN: A Survey

View on arXiv ← Back to list

Authors: Andrew Melnik, Maksim Miasayedzenkau, Dzianis Makarovets, Dzianis Pirshtuk, Eren Akbulut, Dennis Holzmann, Tarek Renusch, Gustav Reichert, Helge Ritter

Published: 2022-12-18 15:04:31+00:00

AI Summary

This survey paper provides a comprehensive overview of state-of-the-art deep learning methods for face generation and editing using StyleGAN architectures. It covers StyleGAN's evolution, relevant metrics, latent representations, GAN inversion techniques, and applications in face editing, stylization, restoration, and deepfakes.

Abstract

Our goal with this survey is to provide an overview of the state of the art deep learning methods for face generation and editing using StyleGAN. The survey covers the evolution of StyleGAN, from PGGAN to StyleGAN3, and explores relevant topics such as suitable metrics for training, different latent representations, GAN inversion to latent spaces of StyleGAN, face image editing, cross-domain face stylization, face restoration, and even Deepfake applications. We aim to provide an entry point into the field for readers that have basic knowledge about the field of deep learning and are looking for an accessible introduction and overview.

Key findings

The survey highlights the advancements in StyleGAN architectures and their successful applications in diverse areas. It identifies key challenges and future research directions, including improvements in temporal consistency, mobile applications, and the integration of StyleGAN with other techniques like diffusion models and NeRFs. The paper also analyzes the trade-offs between reconstruction quality, perceptual quality, and editability in GAN inversion.

Approach

The paper surveys existing methods for face generation and editing using StyleGAN, analyzing different architectures (StyleGAN, StyleGAN2, StyleGAN3), latent spaces (Z, W, W+, S), inversion techniques, and applications in various domains like face restoration and deepfakes. It also explores different loss functions and evaluation metrics.

Datasets

Flickr-Faces-HQ (FFHQ), CelebFaces Attributes Dataset (CelebA), and CelebA-HQ

Model(s)

StyleGAN, StyleGAN2, StyleGAN3, various encoder networks (pSp, e4e, ReStyle), and other models mentioned in relation to specific applications (e.g., GFP-GAN, MyStyle, CLIP)

Author countries

Germany, UNKNOWN

← Previous