New Course:  CS-395 / 495 -26  
IBMR: Image Based Modeling and Rendering

• Until about 6 years ago, computer graphics was mostly an irreversible process that made images from elaborate descriptions of lighting, geometry, surfaces, textures, and movements.   However, many objects are very difficult to describe this way, such as a buttery croissant, a tiger, a cloud, a waterfall, a flickering candle, or a plush toy animal with long, soft hair. Of course, all of these are easy to photograph (assume the tiger is freeze-dried), so why can't we just use the photos instead, somehow?
• How can we get a really good photograph of a cloud if the sun is behind it--what if the sun is to bright and the shadows in the cloud are too dark for any good exposure setting?
• If a single frame of a computer graphics image took 8 hours to render (commonplace for movies such as "Shrek" or "Monsters, Inc."), and the next movie frame is only slightly different, why should we have to render all of it all over again? Can't we re-use most of it somehow, and just render the changes?
• Suppose you have a set of 15 photographs, all taken at the same instant, but taken by different cameras  from different positions, of actor Keanu Reeves in an acrobatic jump-kick.  Could you compute the image from some other, new position? Could you make this 'virtual camera' whirl around the actor (as in the movie "The Matrix"?).  Could you compute these images with radically different lighting? Could you make his clothes shiny, like patent leather or plastic, or give him a beard? Could you give him red hair, freckles, and long braided pigtails that flop around as he moves his head?

Images hold partial information about lighting, objects, surfaces, positions and movements--that's why people like to look at them. This course will introduce you to some recent techniques that try to extract, modify, and/or re-use some of this information to make new images.