Blue-Noise Dithered QMC Hierarchical Russian Roulette

Abstract

In order to efficiently sample specular-diffuse-glossy and glossy-diffuse-glossy transport phenomena, Tokuyoshi and Harada introduced hierarchical Russian roulette, a smart algorithm that allows to compute the minimum of the random numbers associated to leaves of a tree at each internal node. The algorithm is used to efficiently cull the connections between the product set of eye and light vertices belonging to large caches of eye and light subpaths produced through bidirectional path tracing. The original version of the algorithm is entirely based on the generation of semi-stratified pseudo-random numbers. Our paper proposes a novel variant based on deterministic blue-noise dithered Quasi Monte Carlo samples.