Robust Light Transport Simulation via Metropolised Bidirectional Estimators

摘要

Efficiently simulating light transport in various scenes with a singlernalgorithm is a difficult and important problem in computer graphics.rnTwo major issues have been shown to hinder the efficiency ofrnthe existing solutions: light transport due to multiple highly glossyrnor specular interactions, and scenes with complex visibility betweenrnthe camera and light sources. While recent bidirectional path samplingrnmethods such as vertex connection and merging/unified pathrnsampling (VCM/UPS) efficiently deal with highly glossy or specularrntransport, they tend to perform poorly in scenes with complexrnvisibility. On the other hand, Markov chain Monte Carlo (MCMC)rnmethods have been able to show some excellent results in scenesrnwith complex visibility, but they behave unpredictably in scenesrnwith glossy or specular surfaces due to their fundamental issue ofrnsample correlation. In this paper, we show how to fuse the underlyingrnkey ideas behind VCM/UPS and MCMC into a single,rnefficient light transport solution. Our algorithm is specifically designedrnto retain the advantages of both approaches, while alleviatingrntheir limitations. Our experiments show that the algorithm canrnefficiently render scenes with both highly glossy or specular materialsrnand complex visibility, without compromising the performancernin simpler cases.