3. Changes in Research Direction
3.1 Unifying Forward and Inverse Rendering
We plan a new direction in our multi-site telecollaboration project, which also features a collaboration with the University of Pennsylvania. We will attempt to relate and in a sense unify forward rendering (image synthesis) and inverse rendering (geometry extraction). In part, we are considering building an image synthesis and analysis framework around Cornell's Light Measurement Laboratory and their associated software rendering environment. The cornerstone of the approach uses Cornell's already developed capability to generate synthetic images that precisely match photos of the controlled physical environment. With this capability we can use known geometry, lighting, and surface property information to synthesize realistic images of the environment. Conversely, given multiple photographs of the same scene, along with lighting, surface property data, and camera positions, we believe that we should be able to extract the scene geometry within specified accuracy bounds.
Furthermore, we believe that it might be possible to use the extracted geometry with lighting and surface property information to reconstruct scene images using (for example) image-based rendering techniques. Given such a framework we anticipate also being able to evaluate the effectiveness of different geometry-capture approaches under different circumstances. Likewise, we believe we will be able to quantify the results of different scene reconstruction approaches because we will have access to real or realistic (synthetic) images from the controlled environment of the Cornell's Light Measurement Laboratory. Through our association with the University of Pennsylvania, we are also able to take advantage of some the resources of the National Scalable Cluster Project, including high speed networks and supercomputers for exploring the real-time implementation of inverse rendering.
Our approach in computer graphics differs from that of the vision community in its emphasis on using a more comprehensive model of the physics of light interaction to generate a synthetic image that rivals a photograph. This knowledge should give us a powerful vantage point from which to pursue the inverse problem.
3.2 Expansion of Research Groups
Two sites have undergone major expansions this year which have allowed the Center to broaden its research agenda. Caltech added two experienced computer graphics faculty, Drs. Jim Arvo (rendering and inverse rendering) and Peter Schroeder (wavelets and rendering), and an Assistant Director for the Caltech Computer Graphics Lab, Dr. David Breen (telecollaboration and cloth modeling), all with significant research track records. They are now able to augment the research mission for the Center by adding long-term projects in gesture-based interaction, the impact of non-lambertian reflection on inverse problems in radiative transfer, and in wavelet methods for ECG/EEG visualization.
The University of Utah has added three new faculty, Peter Shirley (rendering and visualization), John Hollerbach (haptic interfaces and robotics) and Chuck Hansen (visualization).
Arvo, Shroeder and Shirley are already actively engaged in several new multi-site collaborations, funded in part by the Director's Pool.
Back to the Table of Contents
Forward to Outreach