Events

Talk Gaussian curvature on polyhedral surfaces

22.08.2006 14:00
Informatikzentrum, Seminarraum G30

Speaker(s): Anita Sellent

The Gaussian curvature of a regular surface is a well-known term in differential geometry; there are even several equivalent ways to define it. Gauss himself defined this curvature via the normal image. In the presentation this definition is transferred to polyhedral surfaces: Though the procedure is obvious for convex vertices it causes some difficulties for non-convex vertices. A method to calculate the normal image for some of the non-convex vertices is presented which on the one hand confirms the angle-deficit as polyhedral Gaussian curvature and on the other hand bears further information on the vertices. In conclusion some of the fundamental theorems for the continuous Gaussian curvature can be affirmed for its polyhedral analogue.

Talk Boosting Restricted to Integer Lattices

18.07.2006 14:00
Informatikzentrum, Seminarraum G30

Speaker(s): Dirk Breitenreicher

When classifying a set of samples into various classes in most cases one highly complex optimization problem has to be solved. A boosting approach, in contrast, is able to determine multiple weak learners such that their combination is as accurate as the single, strong classification rule. Often these weak learners are easier to obtain than the solution of the complex optimization problem. However, finding weak learners that can be used for boosting is a not trivial task. Obtaining multiple weak learning rules may also require a lot of time and memory. Classifying on integer lattices can address this problem. When using classifiers on integer lattices each weak learner is build up on integer variables exclusively. This causes that the training and the evaluation time as well as the required memory decreases dramatically in most cases.

Talk Automatic Creation of Object Hierachies for Ray tracing of Dynamic Scenes

26.04.2006 14:00
Informatikzentrum, Seminarraum G30

Speaker(s): Martin Eisemann

Ray tracing acceleration techniques most often consider only static scenes, neglecting the processing time needed to build the acceleration data structure. With the development of interactive ray tracing systems, this reconstruction time becomes serious bottleneck if concerned with dynamic scenes. I am going to describe two strategies for efficient updating of bounding volume hierarchies (BHV) for scenarios with arbitrarily moving objects. The first exploits spatial locality in the object distribution for faster reinsertion of the moved objects. The second allows insertion and deletion of objects at almost constant time by using a hybrid system, which combines benefits from both spatial subdivision and BHVs. Depending on the number of moving objects, the algorithms adjust a dynamic BHV six to one hundred times faster than it would take to rebuild the complete hierarchy, while rendering times of the resulting hierarchy remains almost untouched.