(L) [2015/09/02] [ost
by Paleos] [Combining Path Tracing with MLT] Wayback!The main purpose of using Metropolis Light Transport is for improving efficiency on the paths that are unlikely to be found by chance, such as sharp caustics or light coming in through a crack in a almost closed door. However it is inferior to ordinary path tracing techniques when the path is highly likely to be found. One step taken to restrict MLT to what it is good at, is to use MLT for indirect light only.
A simple way to go further would be to only use it for paths that exceed a brightness threshold.
(L) [2015/09/03] [ost
by friedlinguini] [Combining Path Tracing with MLT] Wayback!I'm not sure thresholding would help. MLT already favors bright paths, and allowing mutations to dim paths would reduce the probability of getting stuck in a small region of path space. I think discontinuities would be noticeable as well.
(L) [2015/09/03] [ost
by bachi] [Combining Path Tracing with MLT] Wayback!>> friedlinguini wrote:I'm not sure thresholding would help. MLT already favors bright paths, and allowing mutations to dim paths would reduce the probability of getting stuck in a small region of path space. I think discontinuities would be noticeable as well.
I think clamping the geometry-BRDF product as in the VPL rendering methods, and applying MLT only on the residuals could be helpful (since these residuals are supposed to be those paths that are not importance sampled well). The "stucking" problem can be alleviated by replica exchange/parallel tempering or something like that. Has someone tried this before?