This talk will survey some of my recent work on estimating the "shape" of stuff in the real world.  I work both on low level vision algorithms for estimating shape and relatively higher level computer graphics methods for building parametrizable and animatable models given raw geometry.  Topics will include:  stereo reconstruction for non-lambertian surfaces, spacetime stereo for acquiring the geometry of moving objects, improving shape estimation by combining photometric stereo and triangulation, and finally constructing an animatable model of the space of all human shapes and poses. Two of these topics were published in computer vision conferences, and two in computer graphics conferences, so I believe the talk will have something to offer both crowds.

James Davis is currently an assistant professor at University of California at Santa Cruz, with research focused primarily on acquiring digital models of the real world for use in computer graphics. He also serves as a technical advisor for VSeeLab, a startup developing video based collaboration tools. After obtaining his PhD from Stanford University in 2002, he spent two years at Honda Research Institute USA, working on real-time range sensing for humanoid robotics applications.