Volume Rendering Architecture Case Study - VolumePro

1
(No Transcript)
2
Volume Rendering ArchitectureCase Study -
VolumePro

• Presentation By
• Boaz Ophir
• Isak Levinson

Visualization and Animation winter 2002
3
Volume Visualization Applications

• Applications
• Medical data
• Seismic data
• Finite element models
• Demands
• Real time
• Interactivity
• Low cost

4
Challenges

• Huge amounts of voxel data
• Data access
• Storage
• Transport
• Computation
• Relatively simple computations per voxel

5
Solution Ideas

• Data Acces
• Smart data partitioning
• Parallel data access
• Computation
• Parallelizm
• Pipelining

6
VolumePro General Overview

• Single chip real time volume rendering system for
  PCs
• System consists of
• PCI card
• Companion 3-D graphics card
• Software
• VolumePro PCI card
• 128MB volume memory
• Vg500 rendering chip (ASIC)

7
VolumePro General Overview (cntd.)

• Chip architecture based on the cube-4 volume
  rendering architecture
• SUNY stony brook
• EM-cube
• Enhanced memory cube-4
• Produced by Mitsubishi electric real time
  visualization group

8
Main Features

• Data
• Rectilinear datasets
• 8/12bit voxels
• Per-sample
• Classification
• Interpolation
• Gradient estimation
• Phong ilumination
• Compositing
• All parameters can be adjusted in RT

9
Main Features (cntd.)

• Classification
• Assigns RGBA values using 4096x36 LUT
• 24 bit precision RGB values
• 12 bit precision a values
• Interpolation
• Tri-linear weighted average of closest voxels
• Phong Ilumination
• Pre-computed reflectance maps for diffuse and
  specular illumination
• Each map sums all illumination from all light
  sources
• Unlimited number of directional light sources
• Opacity and illumination are optionally
  multiplied with gradient magnitude
• Base plane Compositing
• Front-to-back alpha blending
• Minimum/maximum intensity projections

10
System Architecture

• 4 identical rendering pipelines
• 125 MHz
• On-chip voxel distribution network
• Voxel memory
• Pixel memory
• PCI bus
• Block-and-bank skewing scheme

11
System Architecture (cntd.)
12
Rendering Algorithm

• Create 2D image from the volume data set

13
Ray Casting

• Casting the rays through the volume and define
  sample points
• Assigning color and opacity to the sample points
• Interpolating voxels to the new sample points
• Calculate gradients
• Assign lighting to the image
• Accumulate color and opacity to create the image

14
Ray Casting Casting Rays

• Casting the rays through the volume and define
  sample points

15
Ray Casting Casting Rays 2

• Shear-warp matrix

• Handle unisotropic data

• Advantage voxels processed in planes slices

16
Ray Casting Assigning Color

• Assign RGBA value to each interpolated voxel
• RGBA is taken from 4 lookup tables

17
Ray Casting Interpolation
Trilinear interpolation of voxels to the new
sample points
18
Ray Casting Gradients

• Method
• Central differences
• User pre-computed
• Purpose
• Correct unisotropic images
• Used by the illumination model
• Used for color accumulation

19
Ray Casting Illumination

• Phong lighting model
• Calculate gradients
• Diffuse reflection
• Specular reflection
• Ambient reflection
• Emission

20
Ray Casting Accumulating Color

• Accumulate color and opacity to create the image
• Perform alpha correction
• Front to back accumulation

21
Ray Casting Accumulating Color Example
22
Voxel Memory

• Four 16-bit memory interfaces
• Up to 128 MB SDRAM
• Miniblocks
• 2x2x2 neighbouring voxels
• Miniblocks are stored linearly
• Read/written in bursts (SDRAM fast burst mode)
• 3-D skewing of miniblocks
• Distribution across memory modules
• Groups of 4 adjacent MBs (in any direction) are
  stored in separate memory modules

23
Voxel Memory (cntd.)

• Further skewing whithin each memory module so
  adjacent MBs never fall into same memory bank of
  SDRAM chip
• Calculation
• Voxel Coord (u,v,w)
• MiniBlock Coord (um,vm,wm)(u/2,v/2,w/2)
• Module (umvmwm) mod 4
• Bank ((um/4)(vm/4)(wm/4)) mod 4

24
(No Transcript)
25
Voxel Memory (cntd.)

• Voxel memory must be allocated so all dimentions
  are multiples of 32 voxels
• Automatic cropping during rendering
• Arranging voxels in MiniBlocks,
  skewing/deskewing, re-shuffeling and rearranging
  based on view direction are all done by hardware
• Result
• Rendering pipelines always have access to 4
  adjacent miniblocks

26
Advanced Features

• SuperSampling
• In the Z axis
• Supervolumes Subvolumes
• 2563 voxels in one pass
• Software partitions larger volumes and combines
  results
• Multiple Subvolumes can be pre-loaded to volume
  memory
• Subvolumes can be updated in-between frames
• Dynamic and partial updates to achieve animation
  effects
• Loading volume in pieces to pan through volume

27
Advanced Features (cntd.)

• Cropping
• Multiple clipping planes parallel to volume faces
• Combination of intersections, unions inverses

• Cut Planes
• Single cut plane
• Arbitrary orientation and thickness
• Falloff parameter for smooth cuts

• 3-D cursor
• Hardware generated
• Software controlled

28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
VLI Volume Library Interface

• VLI API
• C classes
• Full access to vg500 chip features
• Works cooperatively with 3D graphics libarry
  (such as OpenGL)
• VLI classes
• Data handling
• Voxel storage
• Voxel format
• Transformations shearing, scaling, positioning

32
VLI (cntd.)

• Rendering elements
• Color opacity LUTs
• Lights
• Cut planes
• Cropping
• Etc.
• Rendering context
• Specify data set
• Rendering parameters
• Much much more

33
Future Developments

• Additional features
• Perspective projections
• Intermixing polygons with volume data
• Additional voxel formats
• Improvements
• Increased speed
• Increased memory
• Reduced cost