Scene Relighting

Scene relighting aims to realistically change the illumination of a scene in images or 3D models, decoupling lighting from geometry and material properties. Current research focuses on developing efficient inverse rendering methods, often employing 3D Gaussian representations or neural radiance fields (NeRFs), to reconstruct scenes from multiple images and enable relighting under novel lighting conditions. These advancements leverage techniques like spherical harmonics for shadow enhancement and sun-alignment strategies for simplified outdoor relighting. The ability to accurately and efficiently relight scenes has significant implications for various applications, including virtual staging, game development, and data augmentation for autonomous driving.