In computer graphics, detail is an important part of the big picture. The eye uses fine detail in images to learn about materials, and leaving this structure out lends surfaces a too-smooth appearance that makes them look synthetic. This talk looks at at two projects that focus on getting the details right and find that surprisingly small features matter. The first starts with the straightforward problem of rendering rough surfaces, which are covered with details much smaller than pixels. Techniques for filtering out the too-small details are good at getting the right average color, but we show that they lose glittery, glinty effects that are important to the appearance. We have developed new methods that make it practical to render these materials accurately. The second project addresses textiles, an important category of materials that is difficult to handle with conventional surface rendering. Instead we model textiles not as surfaces but as thin volumes of fibers, using geometric data derived from high-resolution CT scans. This approach produces highly realistic renderings even in closeup, and at the same time, moving complexity from reflectance models to fiber geometry makes the method very general. This is collaborative work with students and colleagues at Cornell, UCSD, RISD, and Autodesk.
Steve Marschner is Professor of Computer Science at Cornell University, where he is conducting research in computer graphics and vision, centered around how optics and mechanics determine the appearance of materials. He obtained his Sc.B. from Brown University in 1993 and his Ph.D. from Cornell in 1998. He held research positions at Microsoft Research and Stanford University before joining Cornell in 2002. In 2010-11 he was a visiting professor at ETH Zürich. He is recipient of the SIGGRAPH Computer Graphics Achievement Award in 2015, a Sloan Research Fellowship in 2006, and co-recipient of a 2003 Technical Academy Award.
Faculty host : Eftychios Sifakis (email@example.com)