Stratified MCMC - Supplementary Material

Adrien Gruson, Rex West and Toshiya Hachisuka

This supplementary document provides an interactive comparison between our method (stratified MCMC) and:

• the techniques presented in Figure 8: PT, PSSMLT, and ERPT
• gradient domain techniques with random replay: GPT, GPSSMLT

All of the renders were generated on a dual-socket Intel Core E5-2683 v4 Broadwell CPU at 2.1 GHz with 32 cores and 32 GB of memory. All comparisons were done using the same rendering time. The path sampler for all methods is path tracing with next event estimation. The error metric are L1, L2, the relative MSE (MRSE), and the mean absolute percentage error (MAPE).

Comparison of SMCMC and techniques presented in Figure 8 (PT, PSSMLT, ERPT)

Comparison of SMCMC and gradient domain techniques (GPT, GPSSMLT) with random replay

Different rendering time of SMCMC