Scientific Visualization is fast becoming a key technology that provides scientists with insights that enable them to steer their numerical simulations towards solving previously unsolvable problems. However, the size of scientific datasets has witnessed exponential growth in the past few years. This sheer size often makes interactive exploration impossible, as only a small portion of data can fit into main memory at a time and the computation cost is often too high to run in real-time. Despite the importance of time-varying datasets, most previous research has focused on the visualization of steady-state data (i.e., data with only a single time step). This project will attack the challenges of large input sizes posed by time-varying data visualization. There are two important and promising research directions towards handling large-scale problems: data compression techniques and out-of-core techniques. This project will develop integrated lossless compression and out-of-core techniques for large time-varying data visualization, including isosurface extraction and direct volume rendering. It will mainly consider the class of irregular-grid volume datasets represented as tetrahedral meshes, which often arises in computational fluid dynamics, partial differential equation solvers, and other fields.
Specifically, the project will develop new lossless compression techniques for vertex coordinates and scalar values for tetrahedral time-varying volume data. It will also develop new out-of-core isosurface extraction and direct volume rendering techniques for tetrahedral time-varying volume data, and integrate the compression and out-of-core visualization techniques together under a unified infrastructure. The expected results would be a collection of new techniques and a unified, proof-of-the-concept visualization system that will minimize the disk space requirement and the visualization rendering time cost. If successful, the system will efficiently support full visualization functionalities (isosurface extraction and volume rendering) for time-varying datasets much larger than can fit in main memory, with performance expected to be independent of the main memory size available.
