11/3/14 Volume Visualiza0on h3p://imgur.com/trjonqk h3p://i.imgur.com/zcjc9kp.jpg Today s Class Grades & Homework feedback on Homework Submission Server Everything except HW4 (didn t get to that yet) & today s reading are entered & uploaded Let me know if something is missing I s0ll need to tweak the formulas & rela0ve %s Readings for Friday on Volume Visualiza0on The Return of the Crayon Exercise! 1
Today s Class Grades & Homework feedback on Homework Submission Server Readings for Friday on Volume Visualiza0on The Return of the Crayon Exercise! Readings for Friday: (pick one) A survey of algorithms for volume visualiza0on, T. Todd Elvins, 1992 2
"... in 10 years, all rendering will be volume rendering" Jim Kajiya at SIGGRAPH '91 Anima0on is cri0cal: from a sta0c 2D image, it is hard to understand 3D informa0on Applica0ons in: Geoscience, astrophysics, chemistry, microscopy, mechanical engineering, non- destruc0ve tes0ng System Requirements: Understandable data representa0on Quick data manipula0on (change parameters) Reasonably fast rendering (see results quickly) This speed was not available in most systems in 1992 Data is on a 3D lagce, with 1 or more values at each grid point Scalar vs. vector datasets Types of data: Density, pressure, temperature, electrosta0c charge, velocity Sources of data: MRI, CT, PET, Sonogram, Laser scan confocal & other microscopes, simula0on, created by- hand Use mul0ple technologies (leverage different advantages) and combine into a single volume! Tricks for vector data Displaying just a 2D slice of the volume Grid of arrows (for vector data) Tracing a streamline/ribbon/par0cle over 0me However, tensor data s0ll challenging 3
Terms Element: Voxel (single constant value) vs. Cell (tri- linearly interpolated from corners) Grid/lagce Cartesian = cubes Regular=rectangular Structured = warped,non- axis- aligned Block structured Unstructured=general space filling polyhedra Hybrid Non- Cartesian (not this paper) Steps in all volume visualiza0on methods Data acquisi0on Slice pre- processing (adjust contrast, etc) Resample/interpolate (as needed) to propor0onal 3D volume/grid Data classifica0on (a.k.a. thresholding) Add external elements (e.g., radia0on treatment plan, etc.) Mapping to geometric or display primi0ves The key step that varies for different volume visualiza?on algorithms Store, manipulate, transform, shade, display to screen Traversal orders: image order (scanline) and object order (front- to- back or back- to- front) Orthographic (be3er for DVR) vs perspec0ve Photorealism? 4
11/3/14 Surface figng (SF), a.k.a. feature extrac0on or isosurfacing Requires threshold choice expensive to change this interac0vely Methods Contour- connec0ng: contours per slice (originally done by hand), use triangles to web between slice contours Opaque cube/cuberille Marching cubes/tetrahedra/dividing cubes Typically faster than DVR Errors in extrac0on (resolu0on issues) lead to false ar0facts in the rendered volume Implicit Surfaces For a sphere: H(x,y,z) = x2 + y2 + z2 r2 If H(x,y,z) = 0, on surface If H(x,y,z) > 0, outside surface If H(x,y,z) < 0, inside surface 5
Marching Cubes Polygoniza0on: extract triangle mesh from signed distance field "Marching Cubes: A High Resolu0on 3D Surface Construc0on Algorithm", Lorensen and Cline, SIGGRAPH '87. Direct volume rendering (DVR), no intermediate primi0ves Methods Ray cas0ng, summing opacity values along the way Integra0on methods Splagng V- buffer rendering Good for: amorphous features like clouds, fluids, gases Disadvantage: must traverse en0re dataset 6
Ray Cas0ng vs. Rendering Pipeline Ray Cas0ng For each pixel For each object Send pixels into the scene Discre0ze first Rendering Pipeline For each triangle For each pixel Project scene to the pixels Discre0ze last Challenges Choosing appropriate threshhold values & Choosing appropriate color & opacity tables Highly dependent on dataset! Examine data, chose ini0al values, visualize, adjust values, repeat Avoid rendering ar0facts/errors that mislead to incorrect medical diagnoses Resolu0on vs. rendering speed vs. accuracy/ errors Future work: paralleliza0on, automate data classifica0on, make real- 0me 7
More figures at the beginning would have helped the explana0on Uneven level of detail for the different methods (esp for a survey paper) Good paper organiza0on Splagng was invented before 1992, and was a key element to the more recent paper rendering the massive david statue scanned point cloud dataset Want to know more about Pixel slice shearing, Pixel image computers What is the rendering 0me? How slow was it? Readings for Friday: (pick one) Hardware- Accelerated Volume Rendering, Pfister et al., from the Visualiza0on Handbook 2004 8
Ray cas0ng Texture slicing Shear- warp Shear- image Splagng Applica0ons: medicine, bio- technology, engineering, oil & gas explora0on, astrophysics, other sciences Lack of real- 0me performance has prevented more widespread adop0on of volumetric visualiza0on Backward mapping (image order) Loop over pixels, do ray cast Forward mapping (object order) Splagng Steps to rendering Interpola0on/Resampling Gradient Es0ma0on (par0al deriva0ve) Central- difference gradient Trickier when the data acquisi0on is anisotropic or sheared Classifica0on Segmenta0on, transfer func0ons (hard to write for non- experts) Shading (e.g. Phong) Composi0ng 9
Advanced Techniques Ray cas0ng: early ray termina0on Space leaping (empty regions): hierarchical spa0al structure, run- length encoding, convex bounding regions Pre- integra0on: mi0gates ar0facts from undersampled high frequency data Cropping or clipping data Polygon & Volume data rendered together Space, 0me, image quality tradeoffs Predic0ons about future of these algorithms based on trends in data collec0on & hardware produc0on Expects reader to have strong background in computer graphics Single value throughout voxel vs. interpola0on with neighboring values 10
Today s Class Grades & Homework feedback on Homework Submission Server Readings for Friday on Volume Visualiza0on The Return of the Crayon Exercise! Crayon Exercise: Volume Visualiza0on Individual Exercise today Pick a volumetric or layered thing that you find interes0ng and about which you know more than the average bear Try to pick something that s not a medical thing, since it seems that s 99% of what volume visualiza0on does It can be silly, it can be a fic0onal thing What interes0ng internal proper0es are important, but not visible from the outside? What spa0al proximity rela0onships between internal components or between internal and external components or rela0ve sizes of internal/external components are important for understanding this thing? Make at least 2 different sketches of volume visualiza0ons that will clearly show both the internal & external shapes of this interes0ng thing Hint: Using mul0ple media (crayons, pencils, markers) might help 11