Selective ray tracing for interactive high-quality shadows

Christian Lauterbach, Qi Mo
, and Dinesh Manocha
University of North Carolina at Chapel Hill

UNC Powerplant with hard shadows
UNC powerplant with soft shadows
Powerplant benchmark: Our selective ray tracing algorithm can render the 12.7M triangle Powerplant model at 16 fps with hard shadows (left) and over 2 fps with soft shadows with 16 light samples (right) running on a NVIDIA GTX 280 GPU.

We present novel algorithms to achieve interactive high-quality illumination with hard and soft shadows in complex models. Our approach combines the efficiency of rasterization-based approaches with the accuracy of a ray tracer. We use conservative image-space bounds to identify only a small subset of the pixels in the rasterized image and perform selective ray tracing on those pixels. The algorithm can handle moving light sources as well as dynamic scenes. In practice, our approach is able to generate high quality hard and
soft shadows on complex models with millions of triangles at almost interactive rates on current high-end GPUs.

Selective Ray Tracing for Interactive High-Quality Shadows
Technical Repor TR09-004, January 2009, UNC Chapel Hill
Video (AVI, DivX)

Quality comparison

Please see the Appendix (PDF) for high-resolution images.

Hard shadows
Hard shadow comparison

Soft shadows
Comparison soft shadows

Related Links

GAMMA Research Group
Ray Tracing Research at GAMMA

This research was supported in part by NVIDIA, RDECOM, NSF, ARO and DARPA.

We would like to thank Brandon Lloyd for his comments and suggestions on earlier drafts, David Luebke for helpful discussions and Shubhabrata Sengupta for providing the city model. Buddha courtesy of the Stanford Scanning Repository, Sibenik cathedral model courtesy Marko Dabrovic and City model courtesy of Aaron Lefohn and Shubho Sengupta.