Progressive Lightcuts for GPU

SIGGRAPH 2012 (talk)

We test our progressive GPU Lightcuts algorithm on a scene with strong secondary lighting and detailed geometry. In the same time, Lightcuts clearly remains superior to path tracing. However, one million virtual point lights (VPLs) cannot capture the illumination on the blinds in full detail. Increasing the number of VPLs improves quality but consumes more memory; further improvements require accumulating multiple rendering iterations. In order to remove the VPL clamping bias, we additionally apply our progressive clamping relaxation scheme. The plot on the right compares the convergence of different clamping relaxation rates, measured against the reference image on the left.


Lightcuts is an attractive rendering algorithm that efficiently handles complex lighting by adaptively approximating the illumination at a point using clusters of virtual point light (VPL) sources. Two of its limitations are the infinite amount of memory required for variance convergence and the bias introduced by VPL contribution clamping. We present a consistent progressive Lightcuts variant, which converges to the correct solution with a bounded memory footprint. This is essential for high quality rendering, especially considering the tight memory budget on the GPU.

Downloads and links

BibTeX reference

  title = {Progressive Lightcuts for GPU},
  author = {Tom{\'{a}}{\v{s}} Davidovi{\v{c}} and Iliyan Georgiev and Philipp Slusallek},
  journal = {ACM SIGGRAPH 2012 Talks},
  year = {2012},
  url = {},
  ISBN = {978-1-4503-1683-5},
  DOI = {10.1145/2343045.2343047}