The Design and Depiction papers contained some very different applications, from creating effective assembly instructions to automatic magazine and web layout and finally two methods for creating effective non-photorealistic renderings.
Designing Step By Step Assembly Instructions
This paper was an interesting collaboration between cognitive psychologists and computer graphics researchers. The goal is to create a method for creating effective assmebly instructions by machine. Much work has been done in mechanical engineering and robotics circles for determining optimal assembly methods based on which parts block other parts from being added to the system. However, this is the first system to marry that knowledge with knowledge about how people follow directions to create step-by-step instructions.
Adaptive Grid Based Document Layout
This was cool evolutionary, if not revolutionary research paper. A group of researchers (mostly from Microsoft, but also from University of Washington) have put together a new layout system that takes a series of different templates and dynamically chooses among them to provide a layout that is: pleasing to the eye, appropriate for the space and aspect ratio of the display device, and transmits the character of the publication.
Most of the examples were from periodicals that have a distinctive look (The New Yorker, New York Times, Vogue, etc.), but it could be applied to just about any company or system that has a defined layout.
For those familiar with SGML (the basis for HTML), HTML, and systems like FrameMaker, the separation of content from presentation style is not much of a stretch, but the specific application in this case uses some interesting techniques to insure that pages are filled and that items look good on every display device.
My only complaint about this talk is that since so much of the research is from Microsoft, I'll expect to see the results in a proprietary form from them with no licensing capabilities....
Suggestive Contours for Conveying Shape
Look at a line drawing from a human artist, then look at a line-style drawing created by a computer using non-photorealisistic rendering (NPR) techniques. Normally they look pretty different, with the computer-rendered version appearing flat and devoid of highlights.
This paper describes a method for adding more useful detail into real-time NPR techniques. The key to their new technique is to add in what they call "suggested contours." Whereas a contour is a line that is added to an NPR rendering when the surface normal is perpendicular to the viewing angle, a suggested contour is a contour that would appear if the view angle were shifted some. Their technique includes a fast mechanisms for determining these suggested contours and for discriminating against lines that would quickly become unstable.