Neural BRDF Representation and Importance Sampling

Controlled capture of real-world material appearance yields tabulated sets highly realistic reflectance data. In practice, however, its high memory footprint requires compressing into a representation that can be used efficiently in rendering while remaining faithful to the original. Previous works encoding often prioritized one these requirements at expense other, by either applying high-fidelity array compression strategies not suited for efficient queries during rendering, or fitting compact analytic model lacks expressiveness. We present neural network-based BRDF data combines high-accuracy reconstruction with practical via built-in interpolation reflectance. encode BRDFs as lightweight networks, and propose training scheme adaptive angular sampling, critical accurate specular highlights. Additionally, we novel approach make our amenable importance sampling: rather than inverting trained learn them more embedding mapped parameters an which sampling is known. evaluate results on isotropic anisotropic from multiple datasets, performance two different models.