“Implementing Lightcuts”

Lightcuts is a new scalable framework for computing illumination in scenes with complex lighting. It handles different types of illumination, including HDR environment maps, sun/sky models, area lights, and indirect illumination and scales well to handle scenes with complex geometry and non-diffuse materials. The core com- ponent is a scalable method for accurate approximating the illumination from many point lights (e.g., thousands or millions).
First a large set of point lights is generated that approximate the exact lighting. Then these point lights are clustered together into a binary tree called the light tree. For each point to be illuminated, a cut through this tree is adaptively computed that satisfies a perceptually-based error threshold. Evaluating the lights and clusters on this cut produces a high-quality approximation of the total illumination. A related technique, called reconstruction cuts, is used to exploit spatial coherence by interpolating illumination components in image regions where they are sufficiently smooth.
The novel components of the lightcuts framework are described in the companion paper, [Walter et al. 2005]. This sketch will concentrate on the software design and algorithms needed to create an efficient implementation of the lightcuts framework.

Walter, B., Fernandez, S., Arbree, A., Bala, K., Donikian, M., and Greenberg, D. P. 2005. Lightcuts: A scalable approach to illumination. In Proceedings of SIGGRAPH 2005, Computer Graphics Proceedings, Annual Conference Series.