Gradient-Domain Bidirectional Path Tracing
Proc. Eurographics Symposium on Rendering, 2015
Marco Manzi | Markus Kettunen | Miika Aittala |
Jaakko Lehtinen |
Frédo Durand |
Matthias Zwicker |
University of Bern |
Aalto University |
Aalto University |
Aalto University, NVIDIA |
MIT CSAIL |
University of Bern |
Abstract
Gradient-domain path tracing has recently been introduced as an efficient realistic image synthesis algorithm. This paper introduces a bidirectional gradient-domain sampler that outperforms traditional bidirectional path tracing often by a factor of two to five in terms of squared error at equal render time. It also improves over unidirectional gradient-domain path tracing in challenging visibility conditions, similarly as conventional bidirectional path tracing improves over its unidirectional counterpart. Our algorithm leverages a novel multiple importance sampling technique and an efficient implementation of a high-quality shift mapping suitable for bidirectional path tracing. We demonstrate the versatility of our approach in several challenging light transport scenarios.
Materials
publication, results and comparisons, slides
Code
Avalilable on the gradient-domain path tracing project page.