A Hierarchical Automatic Stopping Condition for Monte Carlo Global Illumination

Abstract

We introduce a hierarchical image-space method to robustly terminate computations in Monte Carlo image synthesis, independent of image resolution. The technique consists of a robust convergence measure on blocks which are either recursively subdivided or terminated independently, using a criterion which separates signal and noise based on integral estimates from two separate sample sets. The technique can be easily implemented, as the evaluation of the error measure only requires a second framebuffer and a list of non-terminated blocks. Based on the stopping criterion, one can furthermore sample both the image plane as well as the light sources adaptively. Adaptive sampling reduces the number of samples required to gain the same root mean square error to one quarter in some of our test cases.