Title: Stream Processing of Geometry Speaker: Martin Isenburg UC Berkeley Date: Friday Nov 25, 2005 Time 12-13 Venue: Room 3530, HKUST Abstract I will describe how streaming formats support efficient processing of geometric data. Such formats contain tiny bits of additional information that "finalize" previously read data. This indicates which parts of the data have been completely traversed and gives the necessary guarantees to safely process them and de-allocate their corresponding data structures. I will demonstrate an example pipeline where multiple tasks run concurrently on a data set. One module extracts an iso-surface from a regular volume grid and streams it to a second module for topological clean-up. The result is immediately piped to a third module that simplifies it and streams it to a compression module for compact storage or fast transmission. Having streaming formats enables me to simply command-line-pipe these modules together. In on-going work we use our stream paradigm for geometry processing to compute Delaunay triangulations of arbitrary large point clouds using only limited memory resources. I will present some initial results on this.