Exemplars and their (re-) uses

William T. Freeman
AI Lab
MIT

Time and Place

1305 Newell-Simon Hall
Refreshments 3:15 pm
Talk 3:30 pm

Abstract

As computer memory capacities increase, training-based non-parametric methods become feasible for many problems of image analysis and synthesis. I will illustrate this exemplar-based approach applied to several vision and graphics problems: (a) super-resolution (estimating missing high-resolution details); (b) discriminating shading from paint in images; (c) manipulating the drawing style of lines in images; and (d) texture synthesis.
For super-resolution, I will compare parametric and non-parametric approaches. A common characteristic of the non-parametric approaches is the need to amplify the power of the training set by re-using training examples in different contexts.
Joint work with: Egon Pasztor, Owen Carmichael, Josh Tenenbaum, Alyosha Efros.

A short talk before the main talk: Shapetime Photography
We take a sequence of photographs from a stationary stereo camera. From the data, we compute a composite image consisting of the image data from the surface closest to the camera at every pixel. This reveals the 3-d relationships over time by easy-to-interpret occlusion relationships in the composite image. I'll show various examples of these "shapetime photographs".
Joint work with Hao Zhang.

Speaker Biography

William T. Freeman is an Associate Professor of Electrical Engineering and Computer Science at the Artificial Intelligence Laboratory at MIT, joining the faculty in September, 2001. After receiving his doctorate from MIT, he worked for 9 years at Mitsubishi Electric Research Labs (MERL), doing research on machine learning applied to computer vision, Bayesian models of visual perception, and interactive applications of computer vision. He also developed algorithms for electronic photography at Polaroid, and lived in China for one year as a Foreign Expert at the Taiyuan University of Technology, China.
www.ai.mit.edu/people/wtf

Speaker Appointments

For appointments, please contact Yanxi Liu (yanxi@cs.cmu.edu).