280 likes | 461 Views
Large-scale Volume Rendering on a 200 Megapixel Tiled Display Wall. Joerg Meyer jmeyer@uci.edu Department of Electrical Engineering & Computer Science University of California, Irvine August 31, 2014. HIPerWall. 200 Megapixel Tiled Display H ighly I nteractive P arallelized Display Wall.
E N D
Large-scale Volume Rendering on a 200 Megapixel Tiled Display Wall Joerg Meyer jmeyer@uci.edu Department ofElectrical Engineering & Computer Science University of California, Irvine August 31, 2014
HIPerWall • 200 Megapixel Tiled Display • Highly Interactive Parallelized Display Wall
Calit2 • Highlights • Center of GRAVITY (Graphics, Visualization and Imaging Technology) • Interactive Animation Media Lab • Zeiss Center for Microscopy • Integrated Nanosystems Research Facility Director: G.P. Li
A High-Performance Visualization Environment for Collaborative Research • Display Characteristics • 55+ Tiles • 200+ Mega Pixel Resolution • Visualization Cluster • 55+ Dual Processor Render Nodes • 55+ Terabyte of Storage • Network • Gigabit+(Data, Visual, Sync) • OptIPuter Connectivity PIs: Kuester, Jenks, Zender, Sorooshian, Gaudiot
2D Brain Atlas • High-Resolution 2D Display Rhesus Macaque Monkey Brain(1,400 cross-sections,2,666 pixelsper inch) 200 Megapixels
3D Brain Atlas • Distributed 3D Display 800 Megavoxels
3D Brain Atlas • Distributed 3D Display 800 Megavoxels
Distributed Volume Rendering • Based on CGLX framework • Runs on 25 cluster nodes • Synchronized by head node • Interactive control by head node • Local rendering on each node • One node drives two displaysÞ 50 tiles
Distributed Volume Rendering • Uses hierarchical space subdivision (NTFS-based octree) • Multi-resolution decomposition(3D Haar wavelets) • Dynamic viewing frustum clipping on each rendering node • Multi-dimensional transfer functions (3D control widget on head node), Phong-Blinn pre-illumination
High-Resolution Volume Rendering RGB image series (real-color, human brain), 1472 x 1152 x 753, 3.57 GB (Data courtesy of Arthur W. Toga, Dept. of Neurology, UCLA School of Medicine)
High-Resolution Volume Rendering • 3D Reconstruction RGB image series (real-color, human brain), 1472 x 1152 x 753, 3.57 GB (Data courtesy of Arthur W. Toga, Dept. of Neurology, UCLA School of Medicine. Image courtesy of Eric B. Lum, Ikuko Takanashi, UC Davis.)
High-Resolution Volume Rendering • Dynamic Refinement Initial stage Second level of detail Third level of detail Final reconstructed image
High-Resolution Volume Rendering • Dynamic Refinement Close-up Brain
High-Resolution Volume Rendering • Dynamic Refinement Close-up Mandible
High-Resolution Volume Rendering • Dynamic Refinement Skull
Space Decomposition • Dynamic subvolume retrieval on each rendering node • Quick calculation oftree path ... Frustumculling
Combination: Octree/Wavelet • Leaf encoding
Wavelet Compression ... step 2 original (256 x 256) step 1 image pyramid
Original image array Original Volume HLH LLH LLL HLL LL HL L H LHH LHL HLL LH HL HHH First run: x-direction Second run: y-direction Third run: z-direction Wavelet Compression
Distributed Volume Rendering • 3D Texture Buffer Mapping Maximize use of available texture memory
Summary • Interactive, distributed volume rendering • 800 megavoxels • Gigabytedata sets • Dynamicrefinement
Acknowledgements • Huan T. Nguyen, Ph.D.(Univ. of Chicago) • Sebastian Thelen(IRTG) • Prof. Stephen Jenks(UCI) • Sung-Jin Kim, Ph.D.(UCI) • Falko Kuester, Kai-Uwe Doerr(UCSD)
Questions? Joerg Meyer University of California, Irvine EECS/BME Department 644E Engineering Tower Irvine, CA 92697-2625 jmeyer@uci.edu http://imaging.eng.uci.edu/~jmeyer