A statistical framework for comparing importance sampling methods, and an application to rectangular lights

Leonid Pekelis, Christophe Hery

Abstract:

Importance sampling methods for ray tracing are numerous. One reason is due to the lack of a robust, quantitative method for comparison. This tech memo takes strides to provide a framework for comparing sampling methods on fair ground. We define a simple, mathematical notion of strictly preferring one method to another. We demonstrate how to apply the framework in comparing three methods from sampling light rays from a rectangular light source. One method, Angular Stratification, is new to the literature.

Paper (PDF)

Available as Pixar Technical Memo #14-01