Multiresolution View-Dependent Splat Based Volume Rendering of Large Irregular Data Jeremy Meredith, Lawrence Livermore National Laboratory Kwan-Liu Ma, University of California, Davis
Multi-res View-Dependent Splatting Highlights: Visualize very large irregular data interactively Pro: Multi-res: hierarchical oct-tree View-dependent Splatting Con: Oct-tree add much overhead to store volume data
Multi-resolution Hierarchical oct-tree
Splatting Integral of emitted light at each point on image plane
Splatting Low-Albedo Optical Model The amount of emitted light received at location x at the image plane Use Taylor approximation Here, If If(x) is the amount of light at location x in the frame buffer, and αnew(x), Inew(x) are the new incoming opacities and light intensities, respectively OpenGL blending function glBlendFunc(GL SRC ALPHA,GL ONE MINUS SRC ALPHA)
EVL Internal Volume Visualization Workshop Charles Zhang EVL, UIC
Our Goals Very large data Difficulties: WAN, beyond local RAM, interactivity Is multi-res necessary? Pro: fast access, exploration convenient, scalability, pre-fetch compatible Con: a pre-processing must Which mutli-res: wavelet? Is view-dependence necessary? Pro: no redundant data fetch, easy pre-fetching, good for stereo-display(?) Con: overhead? Parallel Volume Rendering Image-ordering or object ordering? IO: accurate OO: fast, suitable for view-dependence Sort-first or sort-last Sort-first uses less bandwidth
Our Goals (cont’d) Interactivity Fast overview? Multi-res, view-dependent? Fast view point change? Tolerance is high when view moving fast. Distributed / Parallel Computing Distributed data on WAN? Is the data scalable, what’s the limit for our viz tools Load-balancing on rendering cluster Multi-tile display The display is scalable?