ReSTIR PT with MCMC Decoupled Shading

We propose a MCMC decoupled shading method that works alongside the original ReSTIR PT process to produce better-converged images without requiring additional rays. By exploiting the structural similarity between ReSTIR PT and MCMC methods, we treat each candidate path generated within ReSTIR PT as if it were spawned using a Markov chain, interpreting spatiotemporal reuses as mutations. This approach allows energy transfer and deposition between pixels, effectively extracting more information from each candidate path within the ReSTIR PT procedure. As a result, our method enhances stratification, reduces correlation, and achieves a more converged output compared to the original ReSTIR PT.