Old News New News Review: Volume Graphics CPSC 314 Computer - PowerPoint PPT Presentation

University of British Columbia Old News New News Review: Volume Graphics CPSC 314 Computer Graphics • extra TA office hours in lab for hw/project • project 4 grading slots signup • for some data, difficult to create polygonal mesh Jan-Apr 2007 Q&A • voxels: discrete representation of 3D object • Wed Apr 18 10-12 Tamara Munzner • next week: Thu 4-6, Fri 10-2 • volume rendering: create 2D image from 3D object • Wed Apr 18 4-6 • translate raw densities into colors and • last week of classes: • Fri Apr 20 10-1 transparencies Blending, Modern Hardware • Mon 2-5, Tue 4-6, Wed 2-4, Thu 4-6, Fri 9-6 • different aspects of the dataset can be emphasized • final review Q&A session via changes in transfer functions Week 12, Mon Apr 2 • Mon Apr 16 10-12 • reminder: no lecture/labs Fri 4/6, Mon 4/9 http://www.ugrad.cs.ubc.ca/~cs314/Vjan2007 2 3 4 Review: Volume Graphics Review: Isosurfaces Review: Isosurface Extraction Review: Marching Cubes • create cube • pros • 2D scalar fields: isolines • array of discrete point 0 1 1 3 2 • classify each voxel 11110100 samples at grid points • formidable technique for data exploration • contour plots, level sets • binary labeling of each voxel to create • 3D array: voxels index • cons • topographic maps 1 3 6 6 3 • use in array storing edge list • find contours • rendering algorithm has high complexity! • 3D scalar fields: isosurfaces • all 256 cases can be derived from 15 base cases 3 7 9 7 3 • closed, continuous • special purpose hardware costly (~$3K-$10K) • interpolate triangle vertex • determined by iso-value • calculate the normal at each cube 2 7 8 6 2 vertex • several methods • render by standard methods • marching cubes is most 1 2 3 4 3 common Iso-value = 5 volumetric human head (CT scan) 5 6 7 8 Review: Transfer Functions To Classify Review: Volume Rendering Algorithms Review: Ray Casting Traversal Schemes Review: Direct Volume Rendering Pipeline • map data value to color and opacity • ray casting • do not compute surface • can be difficult, unintuitive, and slow Intensity • image order, forward viewing Max Classify α α • splatting Average • object order, backward viewing Shade f f • texture mapping Accumulate Interpolate α α First • object order • back-to-front compositing Composite f f Depth 9 10 11 12 Gordon Kindlmann Rendering Pipeline Blending/Compositing Premultiplying Colors • specify opacity with alpha channel: (r,g,b, α ) • how might you combine multiple elements? • α =1: opaque, α =.5: translucent, α =0: transparent • foreground color A , background color B Model/View Geometry Geometry Model/View Perspective Perspective • A over B Lighting Lighting Clipping Clipping Transform. • C = α A + (1- α ) B Database Database Transform. Transform. Transform. Blending • but what if B is also partially transparent ? • C = α A + (1- α ) β B = β B + α A + β B - α β B • γ = β + (1- β ) α = β + α – αβ • 3 multiplies, different equations for alpha vs. RGB Scan Frame- Frame- Scan Depth Depth Texturing Texturing Blending Blending Conversion buffer buffer Conversion Test Test • premultiplying by alpha • C’ = γ C, B’ = β B, A’ = α A • C’ = B’ + A’ - α B’ • γ = β + α – αβ • 1 multiply to find C, same equations for alpha and RGB 13 14 15 16

Reading Rendering Pipeline Modified Pipeline • so far • FCG Chap17 Using Graphics Hardware • vertex shader • replaces model/view, • rendering pipeline as a specific set of stages • especially 17.3 lighting, and perspective with fixed functionality • FCG Section 3.8 Image Capture and Storage • have to implement these • modern graphics hardware more flexible yourself Modern GPU Features • but can also implement • programmable “vertex shaders” replace much more several geometry processing stages • fragment/pixel shader • programmable “fragment/pixel shaders” • replaces texture mapping replace texture mapping stage • fragment shader must do • hardware with these features now called texturing Graphics Processing Unit (GPU) • but can do other things 17 18 19 20 Vertex Shader Motivation Vertex Shaders Vertex Program Properties Vertex Shaders/Programs • hardware transform and lighting: • programmability required for more complicated • run for every vertex, independently • concept effects • i.e. hardware geometry processing • access to all per-vertex properties • programmable pipeline stage • tasks that come before transformation vary widely • was mandated by need for higher • position, color, normal, texture coords, other custom • floating-point operations on 4 vectors properties performance in the late 90s • putting every possible lighting equation in hardware • points, vectors, and colors! is impractical • access to read/write registers for temporary results • previously, geometry processing was done on • replace all of • implementing programmable hardware has • value is reset for every vertex CPU, except for very high end machines • model/view transformation advantages over CPU implementations • cannot pass information from one vertex to the next • downside: now limited functionality due to • better performance due to massively parallel • access to read-only registers • lighting fixed function hardware implementations • global variables like light position, transformation • perspective projection • lower bandwidth requirements (geometry can be matrices cached on GPU) • write output to a specific register for resulting color 21 22 23 24 Vertex Shaders/Programs Vertex Programs Instruction Set Vertex Programs Applications Skinning • arithmetic operations on 4-vectors: • what can they be used for? • a little assembly-style program is executed on every individual vertex • ADD, MUL, MAD, MIN, MAX, DP3, DP4 • can implement all of the stages they replace • want to have natural looking joints on human • operations on scalars • but can allocate resources more dynamically • it sees: and animal limbs • RCP (1/x), RSQ (1/ √ x), EXP, LOG • e.g. transforming a vector by a matrix requires 4 dot • vertex attributes that change per vertex: products • requires deforming geometry, e.g. • specialty instructions • position, color, texture coordinates… • enough memory for 24 matrices • single triangle mesh modeling both upper and • DST (distance: computes length of vector) • registers that are constant for all vertices (changes • can arbitrarily deform objects lower arm • LIT (quadratic falloff term for lighting) are expensive): • procedural freeform deformations • if arm is bent, upper and lower arm remain • lots of other applications • matrices, light position and color, … more or less in the same shape, but transition • shading • very latest generation: • temporary registers • refraction zone at elbow joint needs to deform • loops and conditional jumps • output registers for position, color, tex coords… • … • still more expensive than straightline code 25 26 27 28 Skinning Skinning Skinning Skinning • approach: • arm example: • • in general: • multiple transformation matrices • M1: matrix for upper arm • many different matrices make sense! • more than one model/view matrix stack, e.g. • M2: matrix for lower arm • EA facial animations: up to 70 different matrices (“bones”) • one for model/view matrix for lower arm, and Upper arm: Upper arm: • one for model/view matrix for upper arm • hardware supported: weight for M1=1 weight for M1=1 • every vertex is transformed by both matrices weight for M2=0 weight for M2=0 • number of transformations limited by available Lower arm: Lower arm: registers and max. instruction count of vertex • yields 2 different transformed vertex positions! weight for M1=0 weight for M1=0 programs • use per-vertex blending weights to interpolate weight for M2=1 weight for M2=1 • but dozens are possible today between the two positions Transition zone: Transition zone: weight for M1 between 0..1 weight for M1 between 0..1 Example Example weight for M2 between 0..1 weight for M2 between 0..1 by NVIDIA by NVIDIA 29 30 31 32