Neural Render Proxies for Interactive and Differentiable Lighting
Summary
Disney Research introduces Neural Render Proxies (NRP), a neural-based approach for interactive and differentiable lighting of static scenes. The method decouples path sampling from emission computation, training a lightweight scene-specific network to predict light transport from any location to image pixels, enabling 30–60 Hz relighting with near path-traced fidelity and enabling gradient-based inverse lighting workflows.