In 2008 I worked at North Carolina State University for a research program “Design Tech”. I worked with Abe Stephens from University of Utah and researchers from Nvidia.
Here is the abstract:
Computer graphics rendering creates the imagery widely used in computer games such as Halo and films such as Finding Nemo. These images can be quite expensive to compute, especially when they must be computed in real time for computer games.
Adaptive rendering seeks to expend this computational effort only where it is truly needed. There are a number of adaptive techniques today, but deciding which is best is still difficult. Our project seeks both to develop new adaptive renderers, and new techniques for evaluating adaptive renderers.
With student Mathew Schwartz and collaborators at the University of Utah and NVIDIA Research, PI Watson is creating several short computer animations they will use for adaptive rendering comparison. A “gold standard” animation will form the ideal case, the bar against which they will compare various adaptive renderers. They are also generating several other animations using existing adaptive renders. They will use their new evaluation methods to compare these adaptively rendered animations to an animation created with their new adaptive renderer.
The image above shows four renderings of the same animation with varying quality. Upper right, the lowest quality, one sample for every four pixels, created in 6.5 minutes. Upper left, one sample for every single pixel, created in 14 minutes. Lower left, the highest quality, sixteen samples for every single pixel, created in 170 minutes. Lower right, an adaptive rendering that varies quality throughout the image to maximize quality, while holding render time down to 10 minutes.
All of these animations depict the same scene, in which a car races through an alley. This scene includes dynamic and reflective elements that stress the various adaptive rendering techniques they examine.
Results should enable them to perform more meaningful rendering comparisons and create more effective adaptive renderings, ultimately resulting in better and more convincing imagery in both film and games.