On-line learning of parametric mixture models for light transport simulation

Abstract

Our Our Our BDPT Our -2 TP Our -5 TP Our -30 TP L1 error (abs. difference) Time [minutes]Bidirectional path tracing (BDPT) Our guided BDPT Figure 1: We render a scene featuring difficult visibility with bidirectional path tracing (BDPT) guided by our parametric distributions learned on-line in a number of training passes (TP). The insets show equal-time (1h) comparisons of images obtained with different numbers of training passes. The results reveal that the time spent on additional training passes is quickly amortized by the superior performance of the subsequent guided rendering. AbstractMonte 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.

Real-time online learning of Gaussian mixture model for opacity mapping

“…All of them utilized the batch EM algorithm to train the model. The GMM were also employed to spatially cache and reuse the learned directional distribution on the surfaces [15]. They introduced the online variant of EM [16] to incrementally train the model such that it can serve as the directional importance to globally guide future path samples.…”

“…The weighted stepwise EM [15] generalized the original log-likelihood (Equation 6) to a set of weighted samples…”

Visual importance-based adaptive photon tracing

“…However, they also use an extra pass to trace importons. Vorba et al [19] propose a method to learn the distribution of visual importance in an online training fashion. In our work, we construct the visual importance map in the first eye pass of SPPM.…”

Portal Extraction Based on an Opening Labeling for Ray Tracing

“…Most state of the art methods rely on Path-Tracing and Monte-Carlo estimation to perform the simulation [1,2].…”