SLIDE 1
CSCI 621: Digital Geometry Processing
Hao Li
http://cs621.hao-li.com
1
Spring 2019
Problem Classification
2
13.1 Dynamic Geometry Processing I Hao Li http://cs621.hao-li.com - - PowerPoint PPT Presentation
13.1 Dynamic Geometry Processing I Hao Li http://cs621.hao-li.com - - PowerPoint PPT Presentation
Spring 2019 CSCI 621: Digital Geometry Processing 13.1 Dynamic Geometry Processing I Hao Li http://cs621.hao-li.com 1 Problem Classification 2 Correspondence Classification 3 Correspondence Classification 4 3-D Reconstruction acquisition
SLIDE 2
SLIDE 3
Correspondence Classification
3
SLIDE 4
4
Correspondence Classification
SLIDE 5
5
3-D Reconstruction
acquisition
data provided by Paramount Pictures and Aguru Images
merging registration initial alignment
SLIDE 6
6
Non-Rigid Registration
data provided by Paramount Pictures and Aguru Images
SLIDE 7
7
Full Reconstruction
data provided by Paramount Pictures and Aguru Images
SLIDE 8
8
Correspondence Classification
SLIDE 9
Dynamic Input Data
9
continuous motion / general deformation
SLIDE 10
10
Dynamic Input Data
Data provided with T. Weise and L. Van Gool
SLIDE 11
Dynamic Input Data
11
momentary motion / articulated deformation
SLIDE 12
Animation Reconstruction
12
SLIDE 13
Animation Reconstruction
13
SLIDE 14
Dynamic Shape Reconstruction
SIGGRAPH 2009
SLIDE 15
Template-Based Reconstruction
analyze deformation transfer deformation analyze deformation transfer deformation
SIGGRAPH 2009
SLIDE 16
16
Correspondence Classification
SLIDE 17
Pairwise Correspondence
17
shape & pose / general deformation
SLIDE 18
18
Statistical Shape Spaces
SLIDE 19
19
Statistical Shape Spaces
SLIDE 20
20
Scan Data - Challenges
SLIDE 21
21
Challenges
SLIDE 22
Correspondence Problem
SIGGRAPH 2009
SLIDE 23
Non-Rigid Registration
23
SLIDE 24
target source
Pair of 3D Scans
SIGGRAPH 2009
SLIDE 25
Correspondences are Lost
?
SIGGRAPH 2009
SLIDE 26
Overlapping Regions are Lost
missing data
- verlapping regions
SIGGRAPH 2009
SLIDE 27
Overlapping Regions are Lost
SIGGRAPH 2009
SLIDE 28
Non-Rigid Registration
SIGGRAPH 2009
SLIDE 29
source target registration
detect
- verlap
correspond deform
The Recipe
SLIDE 30
detect
- verlap
correspond deform
deformation ambiguity
The Challenge
SLIDE 31
detect
- verlap
correspond deform
The Challenge
?
SLIDE 32
detect
- verlap
correspond deform
The Challenge
SLIDE 33
detect
- verlap
correspond deform
global optimization via local refinement
helps helps
Observation
SLIDE 34
detect
- verlap
correspond deform
Iterative Global Optimization
SLIDE 35
detect
- verlap
correspond deform closest point pruning global
- ptimization
Robust Non-Rigid ICP
converges? yes relax stiffness no
Iterative Global Optimization
SLIDE 36
detail preservation global consistency
Erigid Esmooth
Deformation Model
de-coupled complexity
SIGGRAPH 2009
SLIDE 37
sparse Cholesky factorization
ci vi
Eplane Etot
+αpoint +αrigid +αsmooth =
Epoint Eplane Erigid Erigid Esmooth
non-linear least squares minimization Gauss-Newton method Jacobian is sparse
that’s it!
Non-Linear Energy Minimization
SIGGRAPH 2009
[Chen & Medioni ’92]
SLIDE 38
38
Summary
Sampling Correspondence Weighting Deformation two scans non-rigid registration
Etot = αfitEfit + αregEreg
In general: Non-linear problem Correspondence must be robust w.r.t. underlying deformation
SLIDE 39
39
Summary
Correspondence Weighting Deformation non-rigid registration Relax Regularization
αsmooth → 0 αrigid → 0
- Example with Embedded Deformation Model
SLIDE 40
40
Symmetries
Source: [Chang and Zwicker 08]
sampling region matching clustering feature matching
SLIDE 41
41
Isometry Preservation
input data sampling correspondence clustering registration
Source: [Huang et al. 08]
SLIDE 42
Dynamic Shape Reconstruction
42
SLIDE 43
Multi-Frame Reconstruction
SIGGRAPH 2009
transfer deformation
SLIDE 44
44
Geometry and Motion Reconstruction
data provided by Stanford and MPI Saarbrücken
SLIDE 45
45
input data template fitting
data provided by Stanford and MPI Saarbrücken
SLIDE 46
More Results
Reconstruction Input Scans Textured Reconstruction
SIGGRAPH 2009
SLIDE 47
More Results
Reconstruction Input Scans Textured Reconstruction
SIGGRAPH 2009
SLIDE 48
More Results
Reconstruction Input Scans Overlaid Scans
SIGGRAPH 2009
SLIDE 49
Template Fitting
49
SLIDE 50
template first scan
Initial Alignment
collaboration with MPI Tübingen/Brown University
SLIDE 51
In Practice: Need Some Correspondences
Collaboration with MPI Tübingen
SLIDE 52
Collaboration with MPI Tübingen
Improving SCAPE
non-rigid alignment
SCAPE model
pose estimation regression
sparse/partial matching accurate model
SLIDE 53
Regression Results
> 50% more accuracy
SLIDE 54
Collaboration with MPI Tübingen
Alignment Comparison
SLIDE 55
Collaboration with MPI Tübingen
Alignment Comparison
SLIDE 56
Template Free-Reconstruction
56
SLIDE 57
Temporally-Coherent Shape Completion
[Li et al. ’11]
partial data reconstruction reconstruction partial data
SLIDE 58
Free-Viewpoint Video
[Li et al. ’11]
SLIDE 59
3D Reconstruction
59
SLIDE 60
Multi-View Capture
multi-view photometric stereo multi-view stereo
SLIDE 61
Single-View Capture
[Newcombe et al. ’11] KinectFusion [Rusinkiewicz et al. ‘02] Artec Group
SLIDE 62
Handling Deformations
[Brown & Rusinkiewicz ’07] [Li et al. ’09] [Chang & Zwicker ’11]
SLIDE 63
Using Human Body Priors
[Tong et al. ‘12] [Cui et al. ‘12] [Weiss et al. ’11]
SLIDE 64
Challenges
low cost deformation, clothing & props daily environment
SLIDE 65
Global Non-Rigid Registration
65
SLIDE 66
3D Scanning
SLIDE 67
Automatic Reconstruction
SLIDE 68
3D Printing
SLIDE 69
http://cs621.hao-li.com
Thanks!
69