The method of multiple sampling by significance for the visualization of functionally defined scenes

Abstract

In ray tracing methods, the key point is to choose the direction for the rays. If many rays are needed not everywhere, but only in some parts of the scene, it is reasonable to increase the selection in these places. As a result, computing resources are not wasted where there is no such need. That is, you need to make selections by significance. In this paper, the visualization of functionally defined scenes is considered. A method of multiple sampling by significance is proposed. The method uses weight functions for multiple sampling by significance. The weighting functions minimize the variance of the multiple sample estimation by significance. Weights can be negative, which reduces the variance. In addition, weights allow you to have additional flexibility when developing a sampling method that accelerates calculations. As a result, acceptable weights were obtained when modeling light transfer. The variance was reduced by using weights in the sample. The dependence of the mean square error on the number of samples is given. Highly realistic functionally defined scenes are visualized. The method is implemented using CPU and GPU. Diagrams of the method's performance are given.