GaRe: Relightable 3D Gaussian Splatting for Outdoor Scenes from Unconstrained Photo Collections

Contributions: โ€ข We propose a novel visibility extraction method by applying binary clustering to the residuals between globally illuminated images and ambient-only rendered counterparts, achieving precise separation of direct sunlight effects.

โ€ข We introduce a region-based supervision framework with a structural consistency loss for physically interpretable and coherent illumination decomposition.

โ€ข We integrate ray tracing into Gaussian representations and design a shadow baking process, achieving real-time relighting with single-pass efficiency.