Variance-Aware Multiple Importance Sampling

Grittmann, Pascal; Georgiev, Iliyan; Slusallek, Philipp; Křivánek, Jaroslav

teaser for Variance-Aware Multiple Importance Sampling

Many existing Monte Carlo methods rely on multiple importance sampling (MIS) to achieve robustness and versatility. Typically, the balance or power heuristics are used, mostly thanks to the seemingly strong guarantees on their variance. We show that these MIS heuristics are oblivious to the effect of certain variance reduction techniques like stratification. This shortcoming is particularly pronounced when unstratified and stratified techniques are combined (e.g., in a bidirectional path tracer). We propose to enhance the balance heuristic by injecting variance estimates of individual techniques, to reduce the variance of the combined estimator in such cases. Our method is simple to implement and introduces little overhead.

In ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia 2019), 38(6): 152:1-152:9, 2019

[teaser] [pdf] [supplemental] [sources] [results] [doi>10.1145/3355089.3356515]

@article{Grittmann2019,
  author          = {Grittmann, Pascal and Georgiev, Iliyan and Slusallek, Philipp and K{\v{r}}iv{\'{a}}nek, Jaroslav},
  title           = {Optimal Multiple Importance Sampling},
  journal         = {ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia 2019)},
  pages           = {152:1--152:9},
  volume          = {38},
  number          = {6},
  year            = 2019,
  month           = nov,
  date            = {2019-11-17/2019-11-20},
  doi             = {10.1145/3355089.3356515},
  publisher       = {ACM}
}