Abstract

Kavita Bala: Making rendering scalable for complex scenes

A fundamental challenge in computer graphics is realistically simulating the appearance of complex natural scenes including effects like fog, indirect illumination, motion blur, and depth-of-field.  Specialized methods exist for rendering these effects, but combining them produces inaccurate images or takes too long. Our insight is that ``more is less'':  the combination of these effects creates new opportunities to eliminate the computation of detail that a human observer cannot perceive.  We exploit this insight for scalable realistic rendering. This research has applications in several areas including cultural heritage and preservation, games and movies, training, architectural planning and engineering design.

I will describe three recent projects: lightcuts/multidimensional lightcuts which introduce multiscale illumination representations coupled with error-bounds;  a matrix formulation of complex rendering that achieves fast performance for previewing applications like cinematic relighting; and finally, new perceptual metrics that predict when appearance is preserved under illumination changes.

Joint work with Adam Arbree, James Ferwerda, Don Greenberg, Milos Hasan, Fabio Pellacini, Ganesh Ramanarayanan, and Bruce Walter.