Visualization Enhancements of Dense Particle Data Sets

James L. BiglerSchool of Computing

University of UtahNovember 8, 2004

Outline

• Motivation and Introduction

• Ambient Occlusion Shading

• Silhouette Edges

Phong Shaded

Data courtesy: Gerald T. Seidler University of Washington

With Silhouettes

With Ambient Occlusion

With Ambient Occlusion and SilhouettesSilhouettes

Why Particle Visualization?

Macro Micro Crop by value

How are Particles Visualized?

Local Lighting Models

Good for local (micro) structure, bad for global (macro) structure.

Shadows

Global Illumination

• Variation in ambient regions

• Soft shadows

• Interreflection of light between surfaces

Wyman Global Illumination for Interactive Isosurfaces

• Wyman et al. cached global illumination values on a grid. Goal was to maintain interactivity during rendering.

Ambient Occlusion or Obscurances

• Zhukov et al. Iones et al.

• Precomputed

• Stored as textures

• Geometric property

Vicinity Shading

• James Stewart• Similar to Wyman, precomputes and stores

in a texture volume for later use in interactive applications.

Silhouette Edges from Depth Buffer

• Usually black, emphasizes view dependent hull of objects

• Saito and Takahashi (“Comprehensible Rendering of 3-D Shapes”)– Cache various aspects of the

rendered image – Use depth and convolution

to find silhouette edges

Particle Ray Tracing

• Parker et al. show in “Interactive ray tracing” that large numbers of particles can interactively be rendered using a parallel ray tracer.

Ambient Occlusion

Texture Resolution

• 16x16 provides a nice compromise– Fidelity– Memory– Computation time

Precomputation Time and Memory

• Using 20 R14K processors on an SGI Origin 3800 (muse.sci.utah.edu). Textures were 16x16 with 49 samples per texel.

955,000

66 min.

233 MB

952,755

261 min.

232 MB

543,088

33 min.

132 MB

7,157,720

12 hours

1,747 MB

Fireball Bullet Foam

Impact on Performance• 10% slower than direct lighting alone.

• However, using only the ambient occlusion values can yield as good as or better performance than direct lighting alone.

Direct lighting

DL with Textures

Textures w/o DL

Fireball 6 16.43 f/s 14.97 f/s 16.75 f/s

Fireball 11 10.55 f/s 9.59 f/s 10.16 f/s

Cylinder 6 13.32 f/s 12.15 f/s 13.37 f/s

Cylinder 22

11.71 f/s 10.94 f/s 11.75 f/s

Bullet 2 28.17 f/s 25.59 f/s 28.79 f/s

Bullet 12 28.76 f/s 25.71 f/s 28.41 f/s

ImagesDirect Lighting

onlyDirect lighting with ambient occlusion

textures

Ambient occlusion textures only

Cylinder 22 Bullet 6 Fireball 11

Impact on Performance• 10% slower than direct lighting alone.

• However, using only the ambient occlusion values can yield as good as or better performance than direct lighting alone.

Direct lighting

DL with Textures

Textures w/o DL

Fireball 6 16.43 f/s 14.97 f/s 16.75 f/s

Fireball 11 10.55 f/s 9.59 f/s 10.16 f/s

Cylinder 6 13.32 f/s 12.15 f/s 13.37 f/s

Cylinder 22

11.71 f/s 10.94 f/s 11.75 f/s

Bullet 2 28.17 f/s 25.59 f/s 28.79 f/s

Bullet 12 28.76 f/s 25.71 f/s 28.41 f/s

Silhouette Edges

• Two options– Precomputation

(object based)– Run time

• Object based

• Image based

Ingredients for Edges

• Image buffer • Depth buffer• Edge detection kernel• Threshold for zero

crossings

-1 -1 -1

-1 8 -1

-1 -1 -1Laplacian kernel

Threshold Edge Response

• Anatomy of a ray

• If a and |b| are the same for each pixel we can use the collection of t as a depth buffer.

Depth Buffer

p(t) = a + tb

t

Performance

Without With

A 17.064 f/s 16.056 f/s

B 2.220 f/s 2.179 f/s

C 2.220 f/s 2.197 f/s

D 1.155 f/s 1.162 f/s

E 2.683 f/s 2.632 f/s

A

B C

D E

Movie

Movie

Questions?