Learning-based human relighting : a survey

Abstract

Human relighting refers to the process of adjusting the lighting effects on human subjects in digital images, 3D scenes, and videos to simulate various lighting scenarios, ultimately achieving realistic visual outcomes. This review provides a comprehensive examination of learning-based human relighting techniques. In doing so, it explores mainstream approaches while systematically documenting the development of related hardware and algorithms. Furthermore, it offers a detailed analysis of how learning-based human relighting methods have evolved across image-based, 3D-based, and video-based contexts. In addition, the review presents an in-depth evaluation of the respective advantages and limitations of these approaches, comparing them across key dimensions such as performance, robustness, and functional capabilities. Finally, it discusses current challenges and future research trends in learning-based human relighting. The goal of this review is to serve as a concise reference guide, offering practical support for both human relighting research and its real-world applications.