Techart Conference

Sanghyun Woo (New York University, USA)

■ Abstract

In this talk, I will discuss my recent project called SwitchLight, which introduces a co-designed approach for human portrait relighting by combining a physics-guided architecture with a pre-training framework. Leveraging the Cook-Torrance reflectance model, we meticulously configured the architecture to accurately simulate light-surface interactions. To address the scarcity of high-quality light stage data, we developed a self-supervised pre-training strategy. This novel combination of precise physical modeling and an expanded training dataset sets a new benchmark in relighting realism.

■ Bio

He is currently a Faculty Fellow of Computer Science at NYU Courant. He received Ph.D. and M.S. degrees in electrical engineering from the Korea Advanced Institute of Science and Technology (KAIST). Prior to that, He earned B.S. degree in electrical computer engineering from Seoul National University (SNU). During his Ph.D. studies, he had the opportunity to intern at Adobe Research (San Jose, CA) and Meta AI (Menlo Park, CA).