Blue Noise Distributed MCMC Decorrelation of ReSTIR

Abstract

Spatiotemporal resampling (ReSTIR) [Bitterli et al., 2020; Lin et al., 2021] is a popular new ray tracing technique. Unfortunately it can suffer from correlation artifacts if left unchecked. One solution for this is offered by Sawhney et al. [2022] in the form of Markov Chain Monte Carlo mutations. We reimplement and evaluate their proposed algorithm, and attempt to optimise it for blue noise. Our addition of blue noise mutations is unsuccessful, but still provides some insight into how the underlying characteristics of decorrelated ReSTIR work against a simple solution for achieving blue noise.