Coherent Metropolis Light Transport with Multiple-Try Mutations

We present in this paper an effective way to implement coherent versions of Metropolis Light Transport (MLT) by using a class of Multiple-Try mutation strategies. Indeed, even if MLT is an unconditionally robust rendering technique which can handle any kind of lighting configurations, it does not exploit any computation coherency. For example, it is difficult to cluster similar light rays into beams or cones, to perform SIMD computations on vectorized data or to efficiently use geometry caching with non-tessellated scenes. To make Metropolis Light Transport suitable to most of the currently existing commercial renderers, we therefore propose to divide the algorithm into two parts: the first one explores the entire integration space in a way very similar to the initial implementation of Metropolis Light Transport while the second one "splits" in an unbiased way each sample into a family of arbitrarily coherent samples. We finally propose to illustrate the efficiency of our approach with an example of implementation of coherent ray tracing using SIMD instructions.