Seminar Announcement
Jack Snoeyink and Jarek RossignacUniversity of North Carolina, Chapel Hill and Georgia Institute of Technology"Immediate Access to Highly Complex, Remote, 4D Models"
Date: Thursday, May 31, 2001
Abstract:Compression and progressive transmisison of meshes of triangles and tetrahedra (J. Rossignac) After developing IBM's "Topological Surgery" technique with G. Taubin, which is now the core of the MPEG-4 standard for 3D compression, Jarek. Rossignac has focused his research efforts, partly supported by the NSF, on a simpler compression scheme, which he called Edgebreaker. Edgebreaker holds the record for the lowest guaranteed upper bound for encoding planar triangle graphs (1.8 bits per triangle). In practice, it compresses the connectivity of triangle meshes down to about a bit per triangle. A new implementation of Edgebreaker, developed in collaboration with A. Szymczak, A. Safonova, H. Lopes, and V. Coors works as a simple state machine. The detailed code fits on three pages and uses, as sole data structure, two arrays of integers, called the Corner Table. The current implementation, which is publicly available, use a parallelogram prediction, but can be easily combined with other schemes for compressing vertex location. Both the Topological Surgery and Edgebreaker have received Best Paper awards. Rossignac and his colleagues have also developed a similar approach, called Grow&Fold, for compressing tegrahedral meshes, such as those used for finite element analysis. Rossignac has also developed progressive techniques for transmitting compressed versions of both triangle meshes and of tegrahedral meshes (with R. Pajarola and A. Szymczak), such that the client may download low resolution models first and then, when necessary, refine them to higher accuracy by downloading compressed upgrades. Isosurfaces and contour trees (J. Snoeyink) Scientific simulations produce data in the form of sample points with intensity values, which are often visualized by drawing level sets or isosurfaces-surfaces of points with the same intensity. One tool that can help in choosing threshold values for interactive exploration of such data is the contour tree, which have been developed by a number of people including van Kreveld, Bajaj, and Pascucci. Contour trees encode the evolution of isosurfaces as the threshold parameter varies. It can be used to compute seed sets for tracing whole or partial isosurfaces, to determine important values for the threshold parameter, and to support flexible contouring. Carr, Snoeyink and Axen developed a simple algorithm to compute a contour tree without computing the set of all isosurfaces, which enables the use of the contour tree as a data analysis tool. Towards a realtime remote visualiztion of time-dependent 3D simulations (J. Rossignac) The engineering and scientific sumulations that are conducted at High-Performance Computing Centers produce hundreds of gigabytes of data, which represent the evolution of several variables over a 4D space-time domain. New compression and progressive transmission techniques are needed to support the interactive exploitation of this data. In collaboration with J. Snoeyink and his colleagues at UNC and with A. Szymczak and S. Menon (Aerospace) at Georgia Tech, Jarek Rossignac has been exploring new approaches for the compression, progressive transmission, and interactive visualization of such data sets. The techniques under consideration view the data as a hyper-terrains in five-dimensions. The natural sampling of the hyper-terrain on a regular 4D space-time grid may be concisely encoded at several levels of resolution using higher-dimensional wavelets or other predictive schemes coupled with variable-length coding. However, in interactive applications where selected subsets of the data must be accessed at specific resolutions, regular representatinos may prove less effective than pentatope meshes produced through an anadptive simplification process or than semi-regular multi-resolution representations. We plan to evaluate all three approaches in the context of a client/server system, where the operator uses two parameters (say pressure P and time T) to interactively control a color-coded iso-surface S(T,P) of all 3D points that take pressure P at time T. The relevance of each iso-surface, measured for example in terms of its surface area, curvature integral, or number of connected components, may be displayed as the hight or color of a P-T control terrain, upon which the (P,T) locations of previously inspected iso-surfaces may be traced and annotated. The global view of the entire dataset provided by the control terrain may help focus interactive inspection on the relevant subsets of the data. We plan to extend to these 4D models the compression and progressive transmission techniques that we have developed for 3D meshes, so that a crude approximation of an initial isosurface S(P,T) may be quickly downloaded and then refined in realtime either to follow user-controlled changes in P or T, or to increase its accuracy during navigation pauses. An interface prototype (J. Snoeyink) We have implemented an initial prototype for the user interaction on the client with Lutz Kettner, a postdoctoral researcher at UNC Chapel Hill. We illustrate this prototype on data sets from combustion simulation, fluid flow, pollution studies, and heat convection in the earth. The data dimensions on the client side are partitioned into a 3D viewer for isosurfaces and a 2D control plane where each point selects a particular isosurface in 3D. Annotations in the control plane help the user to navigate the volume data. We use the number of connected components of the isosurface as an example of an automatically generated annotation that we can compute efficiently using contour trees. We also provide a small pre-computed preview window that is shown continuously at the current cursor position in the control plane. The small preview facilitates a fast overview; selecting the current location in the control plane creates the corresponding detailed view in the 3D viewer for isosurfaces.
Email: Jack Snoeyink, Jarek Rossignac Speaker's web page: Jack Snoeyink, Jarek Rossignac Research web page: Jack Snoeyink, Jarek Rossignac Institution web page: Institution web page: University of North Carolina, Georgia Institute of Technology 
|