1:303:15 Advanced Features - PowerPoint PPT Presentation

1 / 48
About This Presentation
Title:

1:303:15 Advanced Features

Description:

Two levels (bricks and metabricks) Common trick (e.g. Cox and Ellsworth 97) 1 2 3 10 11 12 ... Traversal at finest level of hierarchy can cross brick boundaries ... – PowerPoint PPT presentation

Number of Views:42
Avg rating:3.0/5.0
Slides: 49
Provided by: steve1637
Category:

less

Transcript and Presenter's Notes

Title: 1:303:15 Advanced Features


1
130-315 Advanced Features
  • 130-200 Volume Rendering (Shirley)
  • 200-230 Massive Models (Slusallek)
  • 230-300 Hardware Support
  • GPU ray tracing demo (Foley)
  • Ray processing unit (RPU) (Slusallek)
  • Future challenges (Mark)
  • 315-330 Break

2
Overview
  • Isosurfacing done implicitly at every pixel
  • Care is needed for data layout, but otherwise is
    brute force
  • Direct volume rendering harder but possible

3
Isosurfacing for Analytic Functions
  • f(x,y,z)0
  • ray tracing via root finding (e.g. Kalra and Barr
    89)
  • explicit polygonalization (e.g. Stander and Hart
    97)

4
Trilinear Cells are Easier
5
Isosurfacing for a Trilinear Cell
  • Marching Cubes
  • Lorensen and Cline (87)
  • Wyvill and Wyvill (86)

6
Why Not Always Use Marching Cubes?
  • Marching cubes can generate millions of polygons
  • Reduce by decimation (e.g. Shekhar et. al 96)
  • Reduce by culling (e.g. Livnat and Hansen 98)

7
Isosurfacing for a Trilinear Cell
8
Isosurfacing for a Piecewise Linear Cell
marching cubes
ray tracing
9
Effects of Direct Cubic Solution
10
Effects of Direct Cubic Solution
11
Effects of Direct Cubic Solution
12
Effects of Direct Cubic Solution
13
Effects of Direct Cubic Solution
14
Effects of Direct Cubic Solution
15
Effects of Direct Cubic Solution
16
Effects of Direct Cubic Solution
17
Effects of Direct Cubic Solution
18
Isosurfacing for a grid of cells
ray
19
(No Transcript)
20
Previous Ray Tracing for Isosurfaces
  • Marschner and Lobb (94)
  • Lin and Ching (96)

21
Feature Comparison
  • Ray Tracing
  • Implicit geometry
  • Software shading
  • Marching Cubes
  • Explicit geometry
  • Hardware shading

22
Shadows
without
with
23
Transparency
24
3D textures
25
Optimizations
  • Parallelism
  • Hierarchical data representation
  • Data layout for better locality

26
Hierarchical Data Representation
  • Skip over cells which do not contain an
    isosurface - Wilhelms and van Gelder (90)
  • Keep macrocells which contain the min/max
    values for contained cells

27
Two Level Representation
ray
ray
28
Two Level Representation
ray
ray
29
Two Level Representation
ray
ray
30
Two Level Representation
ray
ray
31
Two Level Representation
ray
ray
32
Two Level Representation
ray
ray
33
Two Level Representation
ray
ray
34
Two Level Representation
ray
ray
35
Number of Hierarchy Levels
  • Traversal from cell to cell is cheaper than
    moving up and down levels
  • Would like to skip large empty regions
  • We use 3 or 4 levels in practice

36
Data Layout (Bricking)
  • Optimizing for memory locality
  • Two levels (bricks and metabricks)
  • Common trick (e.g. Cox and Ellsworth 97)

37
Brick
Metabrick
1 2 3 10 11 12 4 5 6 13 14 15 7 8 9 16
17 18
144...
153...
38
Data Layout (Bricking)
  • Brick sizes (Cache line and page sized cubes)
  • 16 bit data
  • 32 bit data

39
Combining Hierarchy and Bricking
  • Requirements of hierarchy are different than
    the brick sizes
  • Traversal at finest level of hierarchy can cross
    brick boundaries
  • Must compute indices into bricked array

40
Indexing
  • Consider 6x6x6 bricks of 3x3x3 bricks
  • index (x/3/6)666333nynz
    (y/3/6)666333nz (z/3/6)666333
    (x/36)66333 (y/36)6333
    (z/36)333 (x3)33 (y3)3 (z3)
  • Very expensive
  • Integer division and modulus

41
What about that function?
  • index (x/3/6)666333nynz
    (y/3/6)666333nz (z/3/6)666333
    (x/36)66333 (y/36)6333
    (z/36)333 (x3)33 (y3)3 (z3)
  • index fx(x) fy(y) fz(z)

42
Optimization Results
.
0
6
.
8
1
.
5
B
o
n
e


f
r
o
m

f
e
e
t

5
0
.
9
1
1
.
3
1
.
2
Relative times
43
Where time is spent
44
Unstructured data
Mike Parker
45
  • 128 CPU altix (1.3GHz Itanium 2),
  • 1M pixels
  • 10Gb model, 170 timesteps 82x82x332 plus 1.1
    million particles 15-20 fps

James Bigler
46
(No Transcript)
47
Questions
  • What data structure is best for uniform grids?
  • Faster Isosurface Ray Tracing using Implicit
    KD-Trees, Wald, Friedrich, Marmitt, Slusallek and
    Seidel, IEEE TVCG
  • What is best for unstructured grids?
  • Faster direct volume rendering?
  • Custom hardware? (SUNY SB)

48
  • Thanks to
  • Steve Parker
  • Mike Parker
  • James Bigler
  • SGI
Write a Comment
User Comments (0)
About PowerShow.com