Neural-PIL: Neural Pre-Integrated Lighting for Reflectance Decomposition
This addresses a fundamental challenge in computer vision and graphics for realistic scene rendering, offering incremental improvements over existing neural decomposition methods.
The paper tackles the problem of decomposing scenes into shape, reflectance, and illumination from images under varying lighting, proposing a method that replaces costly illumination integrals with a neural network query and learns priors on BRDF and illumination, resulting in more accurate BRDF and light estimates for improved novel view-synthesis and relighting compared to prior art.
Decomposing a scene into its shape, reflectance and illumination is a fundamental problem in computer vision and graphics. Neural approaches such as NeRF have achieved remarkable success in view synthesis, but do not explicitly perform decomposition and instead operate exclusively on radiance (the product of reflectance and illumination). Extensions to NeRF, such as NeRD, can perform decomposition but struggle to accurately recover detailed illumination, thereby significantly limiting realism. We propose a novel reflectance decomposition network that can estimate shape, BRDF, and per-image illumination given a set of object images captured under varying illumination. Our key technique is a novel illumination integration network called Neural-PIL that replaces a costly illumination integral operation in the rendering with a simple network query. In addition, we also learn deep low-dimensional priors on BRDF and illumination representations using novel smooth manifold auto-encoders. Our decompositions can result in considerably better BRDF and light estimates enabling more accurate novel view-synthesis and relighting compared to prior art. Project page: https://markboss.me/publication/2021-neural-pil/