Paul Merrell Dinesh Manocha

1
(No Transcript)
2
Continuous Model Synthesis
Paul Merrell Dinesh
Manocha The
University of North Carolina
at Chapel Hill
3
Model Synthesis

• Procedural Modeling Technique
• Inspired by texture synthesis

Example Texture
Example Model
Output Texture
Output Model
4
Model Synthesis

• Procedural Modeling Technique
• Advantages
• General
• Applies to many types of models
• Simple
• Simple to use, simple to implement
• Efficient
• Generates complex shapes in a few minutes

5
Outline

• Related Work
• Method
• Results
• Comparison

6
Related Work

• Texture Synthesis
• Efros and Leung, 1999
• Wei and Levoy, 2000
• Kwatra et al., 2003
• Model Synthesis
• Merrell, 2007

Axis Aligned Parts
7
Related Work - Procedural Modeling
Wonka et al., 2003
Müller et al., 2006
Legakis et al., 2001
Prusinkiewicz et al., 2001
Mech Prusinkiewicz, 1996
Musgrave et al., 1989
8
How are the Shapes Similar?
Output Shape
Example Shape
9
Create Sets of Parallel Lines
Example Shape
Sets of Parallel Lines
10
Create Vertices at Intersections
Example Shape
2D Arrangement
11
Output is Generated within the Lines
Example Shape
Output Shape
12
Divide the Example with Lines
Example Shape
13
What Could Appear at an Edge?
Possible Edge States
Exterior
Interior
Edge
14
What Could Appear at an Edge?
Possible Edge States
This never appears in the Example Shape.
15
What Could Appear at a Vertex?
Possible Vertex States
Exterior
Interior
Edge
Edge
Vertex
16
What Could Appear at a Vertex?
Possible Vertex States
17
Enumerate All Possibilities Initially
For a triangular example shape
Edges have three states.
Vertices have five states.
Faces have two states.
18
Assign States
We now know something about the adjacent edges.
The edge must be above the blue vertex.
Empty space must be beneath it.
Interior above exterior beneath the edge.
19
Redraw
, Propagate
Each removal may cause additional
removals. The removals propagate.
20
Finished Removing States.
No more states need to be removed.
The result is a triangle.
We havent determined whats here.
Assign more states
Propagate
Repeat
21
Algorithm Overview

• Create sets of parallel lines.
• Find acceptable states.
• while there exists an unassigned edge or vertex
• . Randomly assign a state to it.
• . Remove incompatible states.

Example Shape
22
What Happens in 3D?
Example Model

• Create parallel planes instead of lines.

23
Find All Possible States
Nine States
Exterior
Interior
Vertex
Example Model
Face
Edge
24
Example Model
25
Generated Model
26
Example Model
27
Generated Model
28
Example Model
29
Generated Model
30
Example Model
31
Generated Model
32
Example Model
33
Generated Model
34
Example Model
Sierpinkski Tetrahedron
35
Generated Model
36
Result Statistics

• Total modeling time was short.
• Input models were easily created.
• Each model was generated in two minutes or less.

37
Comparison

• Other methods model complex buildings.
• Müller et al., 2006
• Wonka et al., 2003
• They require more guidance from the user.
• They use shape grammars.
• Earlier work in model synthesis required
  axis-aligned parts.

38
Limitations

• Each distinct normal requires another set of
  planes.
• With many planes, the algorithm becomes less
  efficient.
• More user control is needed.

39
Conclusion

• Model synthesis is a general, simple, and
  efficient method for generating complex models.
• Acknowledgements
• NSF
• ARO
• DARPA
• Intel

40
Constraint-Based Model Synthesis
41
(No Transcript)
42
Modifying in Parts