Variational Tree Synthesiskucg.korea.ac.kr/new/seminar/2016/ppt/ppt-2016-11-24.pdf · Variational...
Transcript of Variational Tree Synthesiskucg.korea.ac.kr/new/seminar/2016/ppt/ppt-2016-11-24.pdf · Variational...
Variational Tree Synthesis
Rui-Wang et al.Computer Graphics Forum 2014
Copyright of figures and other materials in the paper belongs original authors.
Presented by Qi-Meng Zhang
2016. 11. 24
Computer Graphics @ Korea University
Qi-Meng Zhang | 2016. 11. 24 | # 2Computer Graphics @ Korea University
Abstract
• Generate realistic trees in specific shapes
• Formulate the tree modeling as an optimization process
• Applicable to generate trees with different shapes, from interactive design and complex polygonal meshes
1. Introduction
Qi-Meng Zhang | 2016. 11. 24 | # 4Computer Graphics @ Korea University
• Typical methods
L-system
• Modeling trees from symbolic rules
• Capable of generating various types of trees
• Limited at modelling trees under specific global constraints
ex: crown shape
Sketch-based modeling
• From the view angle of user interface
• A trade-off has to be made between the amount of user-involved sketching and the fidelity of resultant models
1. Introduction
Qi-Meng Zhang | 2016. 11. 24 | # 5Computer Graphics @ Korea University
• In this paper, we formulate the shape-guidance tree modelling problem in a multilevel variational optimization framework
Variational cost function is iteratively minimized at each level
• Measures the difference between the guidance shape and the target tree crown
First
∙ Generate initial tree branches
Second
∙ Generate shape sample points and cluster
Finally
∙ Optimize sub-trees to best fit these clustered sample points
1. Introduction
Qi-Meng Zhang | 2016. 11. 24 | # 6Computer Graphics @ Korea University
• Consider the botanical factors constrains in the optimization
Each sub-tree is parameterized in several botanical parameters and constructed
• Including the minimum total branches volume and spatial branches patterns
Why need volume minimization
• Most wildly applied model to depict the optimized structure of tree
1. Introduction
Qi-Meng Zhang | 2016. 11. 24 | # 7Computer Graphics @ Korea University
• Modeling system
Overview
Optimization process
Initialization&Partition Local optimization Global optimization
2. Related Work
Qi-Meng Zhang | 2016. 11. 24 | # 9Computer Graphics @ Korea University
• Extended L-systems
• Sketch-based plant modeling
2. Related work
MECH ˇ R.et al., ”Visual models of plants interacting with their environment” SIGGRAPH 1996
PRUSINKIEWICZ P.et al., ”Synthetic topiary”ACM SIGGRAPH 1994
Boudon F.et al., ”Structure from silhouettes: A new paradigm for fast sketch-based design of trees” Computer Graphics Forum 2009
Longay, S.et al., ”Treesketch: Interactive procedural modeling of trees on a tablet” Eurographics 2012
Qi-Meng Zhang | 2016. 11. 24 | # 10Computer Graphics @ Korea University
• Space colonization method
2. Related work
RUNIONS A.et al., ”Modeling trees with a space colonization algorithm” NPH’ 2007
· Proposed a space colonization methodbut it did not consider the botanicalconstraints
· Improve the problem of space coloniza-tion method, shape constraints optimizedby local branches growth and competition
PALUBICKI W.et al., ”Self-organizing tree models for image synthesis” ACM TOG 2009
Qi-Meng Zhang | 2016. 11. 24 | # 11Computer Graphics @ Korea University
• Tree reconstruction methods
2. Related work
Boris Neubert.et al., ”Approximate Image-Based Tree-Modeling using Particle Flows”ACM TOG 2007
Livny Y.et al., ”Automatic reconstruction of tree skeletal structures from point clouds”ACM TOG 2010
Livny Y.et al., ”Texture-lobes for tree modelling”ACM TOG 2011
·Model trees from acquired data such
as from images, video, scanned pointset .These method tend to accomplish a reconstruction mission rather than a modeling issue.
Qi-Meng Zhang | 2016. 11. 24 | # 12Computer Graphics @ Korea University
2. Related work
RUNIONS A.et al., ”Modeling and visualization of leaf venation patterns” ACM TOG 2005
HAHN H.et al., ”Fractal aspects of three-dimensional vascular constructive optimization”Fractals in Biology and Medicine2005
· Generate branch structure according to
biological factors, result patterns are applicable to texture synthesis
· Proposed a Global Constructive
Optimization method
3. Variational Tree Synthesis Model
Qi-Meng Zhang | 2016. 11. 24 | # 14Computer Graphics @ Korea University
• In a tree structure
Node: An insertion zone of a leaf on the stem
Internode: The stem portion between two successive nodes
Metamer: The basic structural unit of a plant body formed by a node
• An abstract branching model
•
3.1 Preliminaries
,defines N + 1 node ( is root node)
,corresponding spatial positions of the node set
,the diameters of the node set
①
②
③
,the topological relationship ④
Qi-Meng Zhang | 2016. 11. 24 | # 15Computer Graphics @ Korea University
• Pipe model
The diameter of a parent branch has a relationship with its child braches
is a branch transmission coefficient between (2.0,3.0)
• The structure of tree branches
3.2 Botanical Rules
Major morphological features
Qi-Meng Zhang | 2016. 11. 24 | # 16Computer Graphics @ Korea University
The major morphological features of a tree
A tree could represented as a hierarchical structure
• combine all parameter vectors of different levels to be a botanical arguments set
3.2 Botanical Rules
Qi-Meng Zhang | 2016. 11. 24 | # 17Computer Graphics @ Korea University
• The guidance shape
Collect shape nodes to form the constraint set
• Our goal is to approximate the input guidance shape, so there has a
optimization problem
3.3 Variational Model
,defines M+1 nodes
,represent spatial positions
,represent the diameters of corresponding nodes
Qi-Meng Zhang | 2016. 11. 24 | # 18Computer Graphics @ Korea University
• Branch model optimization
• is an object function to present modeling cost
3.3 Variational Model
① ② ③
①
②
③
: the intrinsic structure energy
: the exterior shape-guidance energy
: the parameter control penalty form the botanical parameter set
Qi-Meng Zhang | 2016. 11. 24 | # 19Computer Graphics @ Korea University
• The intrinsic structure energy
Minimize the total volume of branches
• The volume function for a given internode :
is the diameter computed by Equation(2)
is shape coefficient
∙ In general =2.0
3.3 Variational Model
Qi-Meng Zhang | 2016. 11. 24 | # 20Computer Graphics @ Korea University
• The shape-guidance energy
Compute the similarity between the target structure and the shape constraint set
• Distance-aware measurement
• The botanical parameter control penalty energy
3.3 Variational Model
Angle Rotation angle Length of
4. Optimization Algorithm
Qi-Meng Zhang | 2016. 11. 24 | # 22Computer Graphics @ Korea University
• First:
Structure initialization, generated by
• Second:
Shape partition
• Find best guidance shape for each subtree
• Final:
Structure update
4.1 Algorithm Overview
Three-step optimization
Shape partition
Local minima Global optimization
Structure update
Qi-Meng Zhang | 2016. 11. 24 | # 23Computer Graphics @ Korea University
• Recall these botanical parameters are used to create the initial topology and nodes’ positions of branches in the current level of the tree.
4.2 Iterative Optimization at One Level
Structure initialization
Input guidance shape
Structure initialization
Initial topology
Qi-Meng Zhang | 2016. 11. 24 | # 24Computer Graphics @ Korea University
• Recall , and are determined
by branch structure ,so the object function could become
According to the Equation(9), we form an initial partitionof the shape by clustering the shape nodes
4.2 Iterative Optimization at One Level
Shape partition
Using K-Means, clustering sub-tree in k regions, in therecalled k-partition
Has 2 partitions
Qi-Meng Zhang | 2016. 11. 24 | # 25Computer Graphics @ Korea University
• The optimization of internodes will not break the shape-guidance so,
update the branch structure of sub-trees
Minimizes Equation(10) ,use a gradient descent method with numerically gradient evaluations
4.2 Iterative Optimization at One Level
Structure update
Qi-Meng Zhang | 2016. 11. 24 | # 26Computer Graphics @ Korea University
• Given initial K subtrees
• Iteratively taken
Shape partition
• First, all the shape nodes are partitioned into K clusters
Structure update
• Then the resultant branch node position pj is update
4.2 Iterative Optimization at One Level
Major procedure
Two level
Qi-Meng Zhang | 2016. 11. 24 | # 27Computer Graphics @ Korea University
• Jump out local minima
A given node to be the maximal steps connecting shape nodes in
• Branches in smaller order are pruned away from the structure
The corresponding shape sample nodes are reconnected to their nearest branch nodes in the remaining skeletal structure
4.3 Topology refinement
Local minima Global optimization
5. System and Modeling Interface
Qi-Meng Zhang | 2016. 11. 24 | # 29Computer Graphics @ Korea University
• Sketching guidance shape
Sketch silhouettes, the shape of crown is
automatically generated
• Importing mesh
Constrained points(shape nodes)are sampled
from this guidance shape
5. System and Modeling Interface
Input guidance shapes
Importing mesh
Qi-Meng Zhang | 2016. 11. 24 | # 30Computer Graphics @ Korea University
• Shape represented in triangles
Distribute uniform sample nodes on these triangles
• A is the averaged area
• Distribute uniform sample nodes in the shape volume
With same
• Replicate leaves and place them on branches based on botanicallaws governing leaf arrangement
5. System and Modeling Interface
Generation of shape sample nodes & Leaf population
6. Result
Qi-Meng Zhang | 2016. 11. 24 | # 32Computer Graphics @ Korea University
6. results
Modelling abilities
Lateral density Apical growth
Sub-branch Angle
N=5 N=7 N=9 P=0.2 P=0.6 P=1.0
n=1 n=2 n=3 θ=25◦ θ=45◦ θ=65◦
Qi-Meng Zhang | 2016. 11. 24 | # 33Computer Graphics @ Korea University
6. results
Apical and lateral control
P=1.0The apical control be limitedto the main stem
Level 0: P=0.3 N=3Level 1: P=0.5 N=3
Level 0: P=0.2Level 1: P=1.0
Qi-Meng Zhang | 2016. 11. 24 | # 34Computer Graphics @ Korea University
6. results
Alternate vs. opposite lateral
n=1 n=2
Qi-Meng Zhang | 2016. 11. 24 | # 35Computer Graphics @ Korea University
• The topology update can not always guarantee the global decrease of total energy
choose lowest energy
6. results
Local minima and topology refinement
Qi-Meng Zhang | 2016. 11. 24 | # 36Computer Graphics @ Korea University
• Under a same guidance shape
6. results
Comparison & Performance
PRUSINKIEWICZ P.et al., ”Synthetictopiary” ACM SIGGRAPH 1994
PALUBICKI W.et al., ”Self-organizingtree models for image synthesis”ACM TOG 2009
Our method
7. Conclusions and Discussion
Qi-Meng Zhang | 2016. 11. 24 | # 38Computer Graphics @ Korea University
• First
Only consider a small set of hierarchical botanical rules
• Limits our method to model like flowers and vine
• Second
Does not consider the environmental factors
• Ex: gravity or sunshine
• Third
All botanical rules are counted in the optimization by soft constraints
• Some unnatural result
• Finally
Uniformly generate shape sample nodes on triangles
• Final distribution of branches not wanted by users
7. Conclusions and Discussion
Limitations
Qi-Meng Zhang | 2016. 11. 24 | # 39Computer Graphics @ Korea University
• First
Distribute sample nodes according to some botanical patterns with andwithin the shape
• Second
Tree geometry reconstruction
• Integrate with other image based techniques
• Finally
User interface
7. Conclusions and Discussion
Future work