Online path sampling control with progressive spatio-temporal filtering

This work introduces efficient online methods to build all-frequency approximations to the light transport distribution in a scene by exploiting both its spatial and temporal coherence, and shows several ways in which these approximants can be used to control the underlying sampling process and greatly reduce sample variance in a regular path tracer. The left side of the picture shows the approximation built byprogressive spatio-temporal filtering after 32 iterations at 1spp, while the right side shows the converged path traced result employing this approximation as a control variate.