Real-Time Relighting of Human Faces with a Low-Cost Setup

Video-streaming services usually feature post-processing effects to replace the background. However, these often yield inconsistent lighting. Machine-learning-based relighting methods can address this problem, but, at real-time rates, are restricted a low resolution and result in an unrealistic skin appearance. Physically-based rendering techniques require complex models that only be acquired using specialised equipment. Our method is lightweight uses standard smartphone. By correcting imperfections during capture, we extract convincing physically-based model. In combination with suitable acceleration techniques, achieve rates on commodity hardware.