Importance Resampling for BSSRDF

Rendering translucent effects using the diffusion model BSSRDF [Jensen et al. 2001] is still a difficult problem compared to traditional BSDF, because it introduces additional degrees of freedom. In a BSSRDF, the light does not scatter at a single point: it enters the surface at one point, and exits at another. Various techniques have been proposed to sample an exit point based on a entry point. Traditionally, we prebuilt a point cloud [Jensen and Buhler 2002], distributed on the surface of the object, and when rendering we would use it as exit points. However this had multiple problems, for small DMFPs, the density of the point cloud could be arbitrary high, and the heavy precomputation was not pathtracing friendly.