Temporally Adaptive Rendering

By exploiting the coherence in imagery across time, video compression is able to maintain high quality and yet achieve much higher compression rates than can be achieved in compression of single frames. We are exploiting this same coherence in interactive rendering, focusing rendering effort on color edges in space and time. Resulting animations are 10 times faster when maintaining the same fidelity as measured by PSNR. We are currently refactoring our algorithm for the GPU, developing a new method for measuring rendering error along the way.

Project members: Abe Stephens (Utah), Alejandro Fonseca (NCSU), Koushik Naran (NCSU), Abhinav Dayal (NU), Cliff Woolley (UVA), Ben Watson (NCSU), David Luebke (NVIDIA), Steve Parker (NVIDIA/Utah), Pete Shirley (NVIDIA).

Sponsors: NSF award 0093172, NVIDIA Professor Partnership, ATI & Autodesk.

Thanks to: the Manta, OpenRT & BART projects.

Publications

B.A Watson & D. Luebke. (2005). The ultimate display: where will all the pixels come from? IEEE Computer, 38, 8, 54-61. (pdf)

A. Dayal, C. Woolley, B.A. Watson & D. Luebke. (2005). Adaptive frameless rendering. Proc. Eurographics Symposium on Rendering, 265-275. Acceptance rate 33%. (pdf)

A. Dayal, C. Woolley, B.A. Watson & D. Luebke. (2005). Adaptive frameless rendering. Northwestern U., Dept. Computer Science, Technical report NWU-CS-05-07.

A. Dayal, C. Woolley, B.A. Watson & D. Luebke. (2004). Temporally adaptive frameless rendering. 2nd Midwestern Graphics Conference (Evanston, IL).

B.A Watson, D. Luebke, A. Dayal & C. Woolley. (2004). Improving adaptive display with temporally adaptive rendering. Cyberpsychology & Behavior, 7, 6. (pdf)

A. Dayal, C. Woolley, B.A. Watson & D. Luebke (2004). Temporally adaptive frameless rendering (pdf). Technical report NWU-CS-04-47.

A. Dayal & B.A. Watson. (2003). Adaptive frameless rendering. 1st Midwestern Graphics Conference (St. Louis, MO).

A. Dayal, B.A. Watson & D. Luebke (2002). Improving frameless rendering by focusing on change. Technical sketch, ACM SIGGRAPH 2002 Conference Abstracts and Applications, (San Antonio, July), 201. (pdf)

Videos

	Temporally adaptive static scene With fairly static content viewpoint, rendering can rely on old samples in the background, and focus effort on the car. The sampling rate here is 20 times too slow to produce 60 Hz frames. This is a gif animation.
	Head to head comparison Here our rendering of the same scene is compared to an ideal animation and two alternative sampling schemes at the same rate: 60 Hz frames and unbiased frameless sampling. This is a DIVX video.
	Comparison in a dynamic scene A similar comparison, this time with a moving viewpoint. This also is a DIVX video.