For the weekly seminar of the STORM research group, Hugo Rens will give a presentation on Path space analysis and the quest for better samplers in Monte Carlo rendering.

Brief
In the context of rendering, the problem of light transport must be solved, meaning that we want to know how much “light” arrives on a plane representing the image. Many kinds of methods exist to do so, but clearly, Monte Carlo approaches have shown to be quite effective while physically “correct”. These techniques use random samples – drawn in an arbitrary space, from an arbitrary distribution – to construct paths that carry light to the image. The efficiency of such techniques is related to how fast they reduce the error with respect to the number of paths built. The literature provides some intuition on how samples should be distributed, but no solution exists to generate paths directly such that they have good properties. Our goal is to analyze the path space in order to seek for patterns that could be related to accessible scene properties, and in the future, to see if these hypothetical patterns can be exploited to render faster and/or more correctly.
This presentation will be about the problematic of rendering, why Monte Carlo is able to solve it and under which conditions, what problems come with sampling and the various spaces one can actually sample, and, finally, what are the ideas to analyze the path space.

This seminar will take place the 25/05 at 12:30pm in the Salle des Thèses at IRIT.