Jaroslav Křivánek

On-line Learning of Parametric Mixture Models for Light Transport Simulation

Plain Bidirectional path tracing (BDPT)	Our guided BDPT

Comparison of the result of plain bidirectional pah tracing (BDPT) and BDPT guided by our parametric distributions trained in 30 training passes in one-hour rendering of a scene featuring difficult visibility. The results show that the time spent on distributin training is quickly amortized by the superior performance of the subsequent guided rendering.

Abstract

Monte Carlo techniques for light transport simulation rely on importance sampling when constructing light transport paths. Previous work has shown that suitable sampling distributions can be recovered from particles distributed in the scene prior to rendering. We propose to represent the distributions by a parametric mixture model trained in an on-line (i.e. progressive) manner from a potentially infinite stream of particles. This enables recovering good sampling distributions in scenes with complex lighting, where the necessary number of particles may exceed available memory. Using these distributions for sampling scattering directions and light emission significantly improves the performance of state-of-the-art light transport simulation algorithms when dealing with complex lighting.

Publication

Jiří Vorba, Ondřej Karlík, Martin Šik, Tobias Ritschel, and Jaroslav Křivánek.   On-line Learning of Parametric Mixture Models for Light Transport Simulation.   ACM Trans. Graph., SIGGRAPH 2014. ... DOI | BibTeX

Links and Downloads

paper	supplemental document	supplemental images	code	video
pdf (57 MB)	pdf (.4 MB)	pdf (46 MB)	zip (200 MB)	stream

slide notes	slides
pdf (12 MB)	pptx (23 MB)

Image Comparisons

Living room scene	Classroom scene	Door scene
html	html	html

Acknowledgments