A Simplification Architecture for Exploring Navigation Tradeoffs in Mobile VR

1

• A Simplification Architecture for Exploring
  Navigation Tradeoffs in Mobile VR
• Carlos D. Correa
• Ivan Marsic
• Rutgers University
• 2004

2
Abstract

• Application Scenarios
• Context of this research
• Scene graph Simplification
• Problem definition
• Problem transformations
• Video
• Stackable Solvers Architecture
• Experimental Results

3
Application Scenarios
Collaborative Editing
Mobile Augmented Reality
Charles Woodward, VTT Information Technology
Wouter Pasman, Delft University of Technology
Large Dataset Visualization in small devices
4
Interactive vs. Real-time Simplification
Server
Server
Request
Scene
New preferences
Scene
Server
Server
Server
Update
Request
Request
Scene
Delta scene
Delta scene
5
Context
Impostor Generation
Progressive Meshes HLODs Quadric Error
Metrics Image based impostors
Impostor scheduling (combinatorial problem)
Transmission/ Rendering
0.1, 1.0, 0.95, 0.4,
Benefit Heuristics
Progressive transmission MPEG-4 NPR Remote
rendering
Funkhouser and Sequin, 1993 Maciel and Shirley,
1995 Mason and Blake, 2001 Erikson et al., 2001
Simpl. Error metrics User guided simpl. Regions
of Interest
TKP Shaw and Cho, 1998
6
Scene graph Simplification
v1
v1
v1' v1''
v2
v3
v2'
v3'
v2''
v2'
v3
v5
v4
v6
v7
v6'
v7
v5'
v6'
v6''
v6'
v6''
engine
tire
tire
7
Scene Completeness
Scene Completeness
No Scene Completeness
8
Speed-fidelity Tradeoff
DP algorithm R 1000 B 15522 t 3.43 ms
Greedy algorithm R . 1000 B 9096 (58 of
optimal fidelity) t 0.31 ms
9
Problem definition
For each vertex vi define(bi, ri) and (bi', ri
) Let

• SOLVE
• Max ? bi xi ? bi ' yi , (1)
• Subject to
• ? ri xi ? ri' yi ? R (2)
• xi yi ? 1 (3)
• xj yi ? 1 if vi? vj (4)
• xi ? xj yj if vi? vj (5)
• xi, yi 0 or 1 (6)

10
Problem transformations (EMCTKP)
11
Problem transformations (SC)
12
VIDEO
13
Stackable Solvers Architecture
14
Stackable Solvers Architecture (cont)
Application
addNode
setSolution
Optimizer
addNode
setSolution
addNode
Stackable Solver
Transformer
removeNode
setSolution
addNode
setSolution
updateValue
Transformer
setMaxResources
addNode
setSolution
Algorithm
Optimizer
Optimizer
Optimizer
Optimizer
EMCTKP Transformer
EMCTKP Transformer
Filtering
Filtering
EMCTKP Transformer
Partial SC Transformer
SC Transformer
EMCTKP DP Algorithm
TKP DP Algorithm
EMCTKP Greedy Algorithm
TKP DP Algorithm
Suboptimal, No SC, Filtered elements
Exact, Partial SC, Filtered elements
Optimal, No SC
Optimal, SC
15
Fidelity comparison
Scene Completeness

• Benefit ratio benefitgreedy / benefitoptimal
• Example greedy algorithm is 40 optimal for
  xcity44 with R20000.
• Same situation with NO SC greedy algorithm finds
  optimal solution!
• Greedy algorithms are more prone to fail
  (optimality below 50) when
• Scene Completeness
• Scene graph complexity

No Scene Completeness
16
Speed Comparison

• Speed ratio speedgreedy / speedoptimal
• Example greedy algorithm is 2.8 times faster
  than optimal for xcity44 with R1000, but 30
  times faster for R20000.
• For small R, exact algorithm is comparable with
  greedy.
• Exact algorithm computation time increases
  linearly with n and R.

17
Conclusions

• Simplification as Impostor scheduling is a hard
  problem (NP-Complete)
• A variation of TKP has been defined to represent
  the problem
• Choice of algorithm result in speed-fidelity
  tradeoff

• Preferences, e.g. scene completeness, also result
  in navigation tradeoff

• Stackable Solvers Architecture provides a unified
  framework for exploring such tradeoffs and
  enabling mobile VR

18

• Thank You!
• More Info
• http//www.caip.rutgers.edu/disciple
• http//www.caip.rutgers.edu/cdcorrea/research