"I'll present two topics I've been working on recently. The first one
is a global illumination algorithm that targets complex illumination
in occluded environments. It first computes accurate illumination
sparsely in the scene at locations seen from the camera. It then
exploits this cached information in the form of importance to sample
incident illumination from virtual point lights at other locations.
The importance incorporates visibility, allowing for efficient and
robust importance sampling without introducing additional bias to the
solution. It handles direct and indirect illumination simultaneously
and is robust even at sharp discontinuities.
The second topic is about the basics of light path sampling. A new,
mathematically equivalent, formulation of photon mapping in Veach's
the path integral framework reveals exciting new insights in path
sampling robustness and efficiency. It allows us for the first time to
rigorously analyze the photon mapping algorithm and compare it to
bidirectional path tracing. The practical result is a framework that
elegantly combines the two methods after 15 years of independent
co-existance, in order to produce a single unified and efficient
solution to both diffuse and specular light transport. This finally
solves the long-standing mirror-ball-in-a-diffuse-Cornell-box problem
:-)"