PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan...
Transcript of PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan...
![Page 1: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/1.jpg)
MAP Estimation Algorithms in
M. Pawan Kumar, University of Oxford
Pushmeet Kohli, Microsoft Research
Computer Vision - Part I
![Page 2: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/2.jpg)
Aim of the Tutorial
• Description of some successful algorithms
• Computational issues
• Enough details to implement
• Some proofs will be skipped :-(
• But references to them will be given :-)
![Page 3: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/3.jpg)
A Vision ApplicationBinary Image Segmentation
How ?
Cost function Models our knowledge about natural images
Optimize cost function to obtain the segmentation
![Page 4: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/4.jpg)
Object - white, Background - green/grey Graph G = (V,E)
Each vertex corresponds to a pixel
Edges define a 4-neighbourhood grid graph
Assign a label to each vertex from L = {obj,bkg}
A Vision ApplicationBinary Image Segmentation
![Page 5: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/5.jpg)
Graph G = (V,E)
Cost of a labelling f : V L Per Vertex Cost
Cost of label ‘obj’ low Cost of label ‘bkg’ high
Object - white, Background - green/grey
A Vision ApplicationBinary Image Segmentation
![Page 6: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/6.jpg)
Graph G = (V,E)
Cost of a labelling f : V L
Cost of label ‘obj’ high Cost of label ‘bkg’ low
Per Vertex Cost
UNARY COST
Object - white, Background - green/grey
A Vision ApplicationBinary Image Segmentation
![Page 7: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/7.jpg)
Graph G = (V,E)
Cost of a labelling f : V L Per Edge Cost
Cost of same label low
Cost of different labels high
Object - white, Background - green/grey
A Vision ApplicationBinary Image Segmentation
![Page 8: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/8.jpg)
Graph G = (V,E)
Cost of a labelling f : V L
Cost of same label high
Cost of different labels low
Per Edge Cost
PAIRWISE
COST
Object - white, Background - green/grey
A Vision ApplicationBinary Image Segmentation
![Page 9: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/9.jpg)
Graph G = (V,E)
Problem: Find the labelling with minimum cost f*
Object - white, Background - green/grey
A Vision ApplicationBinary Image Segmentation
![Page 10: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/10.jpg)
Graph G = (V,E)
Problem: Find the labelling with minimum cost f*
A Vision ApplicationBinary Image Segmentation
![Page 11: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/11.jpg)
Another Vision ApplicationObject Detection using Parts-based Models
How ?
Once again, by defining a good cost function
![Page 12: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/12.jpg)
H T
L1
Each vertex corresponds to a part - ‘Head’, ‘Torso’, ‘Legs’
1
Edges define a TREE
Assign a label to each vertex from L = {positions}
Graph G = (V,E)
L2 L3 L4
Another Vision ApplicationObject Detection using Parts-based Models
![Page 13: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/13.jpg)
2
Each vertex corresponds to a part - ‘Head’, ‘Torso’, ‘Legs’
Assign a label to each vertex from L = {positions}
Graph G = (V,E)
Edges define a TREE
H T
L1 L2 L3 L4
Another Vision ApplicationObject Detection using Parts-based Models
![Page 14: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/14.jpg)
3
Each vertex corresponds to a part - ‘Head’, ‘Torso’, ‘Legs’
Assign a label to each vertex from L = {positions}
Graph G = (V,E)
Edges define a TREE
H T
L1 L2 L3 L4
Another Vision ApplicationObject Detection using Parts-based Models
![Page 15: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/15.jpg)
Cost of a labelling f : V L
Unary cost : How well does part match image patch?
Pairwise cost : Encourages valid configurations
Find best labelling f*
Graph G = (V,E)
3 H T
L1 L2 L3 L4
Another Vision ApplicationObject Detection using Parts-based Models
![Page 16: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/16.jpg)
Cost of a labelling f : V L
Unary cost : How well does part match image patch?
Pairwise cost : Encourages valid configurations
Find best labelling f*
Graph G = (V,E)
3 H T
L1 L2 L3 L4
Another Vision ApplicationObject Detection using Parts-based Models
![Page 17: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/17.jpg)
Yet Another Vision ApplicationStereo Correspondence
Disparity Map
How ?
Minimizing a cost function
![Page 18: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/18.jpg)
Yet Another Vision ApplicationStereo Correspondence
Graph G = (V,E)
Vertex corresponds to a pixel
Edges define grid graph
L = {disparities}
![Page 19: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/19.jpg)
Yet Another Vision ApplicationStereo Correspondence
Cost of labelling f :
Unary cost + Pairwise Cost
Find minimum cost f*
![Page 20: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/20.jpg)
The General Problem
b
a
e
d
c
f
Graph G = ( V, E )
Discrete label set L = {1,2,…,h}
Assign a label to each vertexf: V L
1
1 2
2 2
3
Cost of a labelling Q(f)
Unary Cost Pairwise Cost
Find f* = arg min Q(f)
![Page 21: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/21.jpg)
Outline
• Problem Formulation
– Energy Function
– MAP Estimation
– Computing min-marginals
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
![Page 22: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/22.jpg)
Energy Function
Va Vb Vc Vd
Label l0
Label l1
Da Db Dc Dd
Random Variables V = {Va, Vb, ….}
Labels L = {l0, l1, ….} Data D
Labelling f: {a, b, …. } {0,1, …}
![Page 23: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/23.jpg)
Energy Function
Va Vb Vc Vd
Da Db Dc Dd
Q(f) = ∑a a;f(a)
Unary Potential
2
5
4
2
6
3
3
7Label l0
Label l1
Easy to minimize
Neighbourhood
![Page 24: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/24.jpg)
Energy Function
Va Vb Vc Vd
Da Db Dc Dd
E : (a,b) E iff Va and Vb are neighbours
E = { (a,b) , (b,c) , (c,d) }
2
5
4
2
6
3
3
7Label l0
Label l1
![Page 25: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/25.jpg)
Energy Function
Va Vb Vc Vd
Da Db Dc Dd
+∑(a,b) ab;f(a)f(b)
Pairwise Potential
0
1 1
0
0
2
1
1
4 1
0
3
2
5
4
2
6
3
3
7Label l0
Label l1
Q(f) = ∑a a;f(a)
![Page 26: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/26.jpg)
Energy Function
Va Vb Vc Vd
Da Db Dc Dd
0
1 1
0
0
2
1
1
4 1
0
3
Parameter
2
5
4
2
6
3
3
7Label l0
Label l1
+∑(a,b) ab;f(a)f(b)Q(f; ) = ∑a a;f(a)
![Page 27: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/27.jpg)
Outline
• Problem Formulation
– Energy Function
– MAP Estimation
– Computing min-marginals
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
![Page 28: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/28.jpg)
MAP Estimation
Va Vb Vc Vd
2
5
4
2
6
3
3
7
0
1 1
0
0
2
1
1
4 1
0
3
Q(f; ) = ∑a a;f(a) + ∑(a,b) ab;f(a)f(b)
Label l0
Label l1
![Page 29: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/29.jpg)
MAP Estimation
Va Vb Vc Vd
2
5
4
2
6
3
3
7
0
1 1
0
0
2
1
1
4 1
0
3
Q(f; ) = ∑a a;f(a) + ∑(a,b) ab;f(a)f(b)
2 + 1 + 2 + 1 + 3 + 1 + 3 = 13
Label l0
Label l1
![Page 30: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/30.jpg)
MAP Estimation
Va Vb Vc Vd
2
5
4
2
6
3
3
7
0
1 1
0
0
2
1
1
4 1
0
3
Q(f; ) = ∑a a;f(a) + ∑(a,b) ab;f(a)f(b)
Label l0
Label l1
![Page 31: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/31.jpg)
MAP Estimation
Va Vb Vc Vd
2
5
4
2
6
3
3
7
0
1 1
0
0
2
1
1
4 1
0
3
Q(f; ) = ∑a a;f(a) + ∑(a,b) ab;f(a)f(b)
5 + 1 + 4 + 0 + 6 + 4 + 7 = 27
Label l0
Label l1
![Page 32: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/32.jpg)
MAP Estimation
Va Vb Vc Vd
2
5
4
2
6
3
3
7
0
1 1
0
0
2
1
1
4 1
0
3
Q(f; ) = ∑a a;f(a) + ∑(a,b) ab;f(a)f(b)
f* = arg min Q(f; )
q* = min Q(f; ) = Q(f*; )
Label l0
Label l1
![Page 33: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/33.jpg)
MAP Estimation
f(a) f(b) f(c) f(d) Q(f; )
0 0 0 0 18
0 0 0 1 15
0 0 1 0 27
0 0 1 1 20
0 1 0 0 22
0 1 0 1 19
0 1 1 0 27
0 1 1 1 20
16 possible labellings
f(a) f(b) f(c) f(d) Q(f; )
1 0 0 0 16
1 0 0 1 13
1 0 1 0 25
1 0 1 1 18
1 1 0 0 18
1 1 0 1 15
1 1 1 0 23
1 1 1 1 16
f* = {1, 0, 0, 1}
q* = 13
![Page 34: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/34.jpg)
Computational Complexity
Segmentation
2|V|
|V| = number of pixels ≈ 320 * 480 = 153600
![Page 35: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/35.jpg)
Computational Complexity
|L| = number of pixels ≈ 153600
Detection
|L||V|
![Page 36: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/36.jpg)
Computational Complexity
|V| = number of pixels ≈ 153600
Stereo
|L||V|
Can we do better than brute-force?
MAP Estimation is NP-hard !!
![Page 37: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/37.jpg)
Computational Complexity
|V| = number of pixels ≈ 153600
Stereo
|L||V|
Exact algorithms do exist for special cases
Good approximate algorithms for general case
But first … two important definitions
![Page 38: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/38.jpg)
Outline
• Problem Formulation
– Energy Function
– MAP Estimation
– Computing min-marginals
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
![Page 39: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/39.jpg)
Min-Marginals
Va Vb Vc Vd
2
5
4
2
6
3
3
7
0
1 1
0
0
2
1
1
4 1
0
3
f* = arg min Q(f; ) such that f(a) = i
Min-marginal qa;i
Label l0
Label l1
Not a marginal (no summation)
![Page 40: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/40.jpg)
Min-Marginals
16 possible labellings qa;0 = 15
f(a) f(b) f(c) f(d) Q(f; )
0 0 0 0 18
0 0 0 1 15
0 0 1 0 27
0 0 1 1 20
0 1 0 0 22
0 1 0 1 19
0 1 1 0 27
0 1 1 1 20
f(a) f(b) f(c) f(d) Q(f; )
1 0 0 0 16
1 0 0 1 13
1 0 1 0 25
1 0 1 1 18
1 1 0 0 18
1 1 0 1 15
1 1 1 0 23
1 1 1 1 16
![Page 41: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/41.jpg)
Min-Marginals
16 possible labellings qa;1 = 13
f(a) f(b) f(c) f(d) Q(f; )
1 0 0 0 16
1 0 0 1 13
1 0 1 0 25
1 0 1 1 18
1 1 0 0 18
1 1 0 1 15
1 1 1 0 23
1 1 1 1 16
f(a) f(b) f(c) f(d) Q(f; )
0 0 0 0 18
0 0 0 1 15
0 0 1 0 27
0 0 1 1 20
0 1 0 0 22
0 1 0 1 19
0 1 1 0 27
0 1 1 1 20
![Page 42: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/42.jpg)
Min-Marginals and MAP
• Minimum min-marginal of any variable =
energy of MAP labelling
minf Q(f; ) such that f(a) = i
qa;i mini
mini ( )
Va has to take one label
minf Q(f; )
![Page 43: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/43.jpg)
Summary
MAP Estimation
f* = arg min Q(f; )
Q(f; ) = ∑a a;f(a) + ∑(a,b) ab;f(a)f(b)
Min-marginals
qa;i = min Q(f; ) s.t. f(a) = i
Energy Function
![Page 44: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/44.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
![Page 45: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/45.jpg)
Reparameterization
Va Vb
2
5
4
2
0
1 1
0
f(a) f(b) Q(f; )
0 0 7
0 1 10
1 0 5
1 1 6
2 +
2 +
- 2
- 2
Add a constant to all a;i
Subtract that constant from all b;k
![Page 46: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/46.jpg)
Reparameterization
f(a) f(b) Q(f; )
0 0 7 + 2 - 2
0 1 10 + 2 - 2
1 0 5 + 2 - 2
1 1 6 + 2 - 2
Add a constant to all a;i
Subtract that constant from all b;k
Q(f; ’) = Q(f; )
Va Vb
2
5
4
2
0
0
2 +
2 +
- 2
- 2
1 1
![Page 47: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/47.jpg)
Reparameterization
Va Vb
2
5
4
2
0
1 1
0
f(a) f(b) Q(f; )
0 0 7
0 1 10
1 0 5
1 1 6
- 3 + 3
Add a constant to one b;k
Subtract that constant from ab;ik for all ‘i’
- 3
![Page 48: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/48.jpg)
Reparameterization
Va Vb
2
5
4
2
0
1 1
0
f(a) f(b) Q(f; )
0 0 7
0 1 10 - 3 + 3
1 0 5
1 1 6 - 3 + 3
- 3 + 3
- 3
Q(f; ’) = Q(f; )
Add a constant to one b;k
Subtract that constant from ab;ik for all ‘i’
![Page 49: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/49.jpg)
Reparameterization
Va Vb
2
5
4
2
3 1
0
1
2
Va Vb
2
5
4
2
3 1
1
0
1
- 2
- 2
- 2 + 2+ 1
+ 1
+ 1
- 1
Va Vb
2
5
4
2
3 1
2
1
0 - 4 + 4
- 4
- 4
’a;i = a;i ’b;k = b;k
’ab;ik = ab;ik
+ Mab;k
- Mab;k
+ Mba;i
- Mba;i
Q(f; ’)
= Q(f; )
![Page 50: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/50.jpg)
Reparameterization
Q(f; ’) = Q(f; ), for all f
’ is a reparameterization of , iff
’
’b;k = b;k
’a;i = a;i
’ab;ik = ab;ik
+ Mab;k
- Mab;k
+ Mba;i
- Mba;i
Equivalently Kolmogorov, PAMI, 2006
Va Vb
2
5
4
2
0
0
2 +
2 +
- 2
- 2
1 1
![Page 51: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/51.jpg)
Recap
MAP Estimation
f* = arg min Q(f; )
Q(f; ) = ∑a a;f(a) + ∑(a,b) ab;f(a)f(b)
Min-marginals
qa;i = min Q(f; ) s.t. f(a) = i
Q(f; ’) = Q(f; ), for all f ’
Reparameterization
![Page 52: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/52.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
– Exact MAP for Chains and Trees
– Approximate MAP for general graphs
– Computational Issues and Theoretical Properties
• Tree-reweighted Message Passing
![Page 53: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/53.jpg)
Belief Propagation
• Belief Propagation gives exact MAP for chains
• Remember, some MAP problems are easy
• Exact MAP for trees
• Clever Reparameterization
![Page 54: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/54.jpg)
Two Variables
Va Vb
2
5 2
1
0
Va Vb
2
5
40
1
Choose the right constant ’b;k = qb;k
Add a constant to one b;k
Subtract that constant from ab;ik for all ‘i’
![Page 55: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/55.jpg)
Va Vb
2
5 2
1
0
Va Vb
2
5
40
1
Choose the right constant ’b;k = qb;k
a;0 + ab;00 = 5 + 0
a;1 + ab;10 = 2 + 1minMab;0 =
Two Variables
![Page 56: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/56.jpg)
Va Vb
2
5 5
-2
-3
Va Vb
2
5
40
1
Choose the right constant ’b;k = qb;k
Two Variables
![Page 57: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/57.jpg)
Va Vb
2
5 5
-2
-3
Va Vb
2
5
40
1
Choose the right constant ’b;k = qb;k
f(a) = 1
’b;0 = qb;0
Two Variables
Potentials along the red path add up to 0
![Page 58: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/58.jpg)
Va Vb
2
5 5
-2
-3
Va Vb
2
5
40
1
Choose the right constant ’b;k = qb;k
a;0 + ab;01 = 5 + 1
a;1 + ab;11 = 2 + 0minMab;1 =
Two Variables
![Page 59: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/59.jpg)
Va Vb
2
5 5
-2
-3
Va Vb
2
5
6-2
-1
Choose the right constant ’b;k = qb;k
f(a) = 1
’b;0 = qb;0
f(a) = 1
’b;1 = qb;1
Minimum of min-marginals = MAP estimate
Two Variables
![Page 60: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/60.jpg)
Va Vb
2
5 5
-2
-3
Va Vb
2
5
6-2
-1
Choose the right constant ’b;k = qb;k
f(a) = 1
’b;0 = qb;0
f(a) = 1
’b;1 = qb;1
f*(b) = 0 f*(a) = 1
Two Variables
![Page 61: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/61.jpg)
Va Vb
2
5 5
-2
-3
Va Vb
2
5
6-2
-1
Choose the right constant ’b;k = qb;k
f(a) = 1
’b;0 = qb;0
f(a) = 1
’b;1 = qb;1
We get all the min-marginals of Vb
Two Variables
![Page 62: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/62.jpg)
Recap
We only need to know two sets of equations
General form of Reparameterization
’a;i = a;i
’ab;ik = ab;ik
+ Mab;k
- Mab;k
+ Mba;i
- Mba;i
’b;k = b;k
Reparameterization of (a,b) in Belief Propagation
Mab;k = mini { a;i + ab;ik }
Mba;i = 0
![Page 63: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/63.jpg)
Three Variables
Va Vb
2
5 2
1
0
Vc
4 60
1
0
1
3
2 3
Reparameterize the edge (a,b) as before
l0
l1
![Page 64: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/64.jpg)
Va Vb
2
5 5-3
Vc
6 60
1
-2
3
Reparameterize the edge (a,b) as before
f(a) = 1
f(a) = 1
-2 -1 2 3
Three Variables
l0
l1
![Page 65: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/65.jpg)
Va Vb
2
5 5-3
Vc
6 60
1
-2
3
Reparameterize the edge (a,b) as before
f(a) = 1
f(a) = 1
Potentials along the red path add up to 0
-2 -1 2 3
Three Variables
l0
l1
![Page 66: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/66.jpg)
Va Vb
2
5 5-3
Vc
6 60
1
-2
3
Reparameterize the edge (b,c) as before
f(a) = 1
f(a) = 1
Potentials along the red path add up to 0
-2 -1 2 3
Three Variables
l0
l1
![Page 67: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/67.jpg)
Va Vb
2
5 5-3
Vc
6 12-6
-5
-2
9
Reparameterize the edge (b,c) as before
f(a) = 1
f(a) = 1
Potentials along the red path add up to 0
f(b) = 1
f(b) = 0
-2 -1 -4 -3
Three Variables
l0
l1
![Page 68: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/68.jpg)
Va Vb
2
5 5-3
Vc
6 12-6
-5
-2
9
Reparameterize the edge (b,c) as before
f(a) = 1
f(a) = 1
Potentials along the red path add up to 0
f(b) = 1
f(b) = 0
qc;0
qc;1-2 -1 -4 -3
Three Variables
l0
l1
![Page 69: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/69.jpg)
Va Vb
2
5 5-3
Vc
6 12-6
-5
-2
9
f(a) = 1
f(a) = 1
f(b) = 1
f(b) = 0
qc;0
qc;1
f*(c) = 0 f*(b) = 0 f*(a) = 1
Generalizes to any length chain
-2 -1 -4 -3
Three Variables
l0
l1
![Page 70: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/70.jpg)
Va Vb
2
5 5-3
Vc
6 12-6
-5
-2
9
f(a) = 1
f(a) = 1
f(b) = 1
f(b) = 0
qc;0
qc;1
f*(c) = 0 f*(b) = 0 f*(a) = 1
Only Dynamic Programming
-2 -1 -4 -3
Three Variables
l0
l1
![Page 71: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/71.jpg)
Why Dynamic Programming?
3 variables 2 variables + book-keeping
n variables (n-1) variables + book-keeping
Start from left, go to right
Reparameterize current edge (a,b)
Mab;k = mini { a;i + ab;ik }
’ab;ik = ab;ik+ Mab;k - Mab;k’b;k = b;k
Repeat
![Page 72: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/72.jpg)
Why Dynamic Programming?
Start from left, go to right
Reparameterize current edge (a,b)
Mab;k = mini { a;i + ab;ik }
’ab;ik = ab;ik+ Mab;k - Mab;k’b;k = b;k
Repeat
Messages Message Passing
Why stop at dynamic programming?
![Page 73: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/73.jpg)
Va Vb
2
5 5-3
Vc
6 12-6
-5
-2
9
Reparameterize the edge (c,b) as before
-2 -1 -4 -3
Three Variables
l0
l1
![Page 74: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/74.jpg)
Va Vb
2
5 9-3
Vc
11 12-11
-9
-2
9
Reparameterize the edge (c,b) as before
-2 -1 -9 -7
’b;i = qb;i
Three Variables
l0
l1
![Page 75: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/75.jpg)
Va Vb
2
5 9-3
Vc
11 12-11
-9
-2
9
Reparameterize the edge (b,a) as before
-2 -1 -9 -7
Three Variables
l0
l1
![Page 76: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/76.jpg)
Va Vb
9
11 9-9
Vc
11 12-11
-9
-9
9
Reparameterize the edge (b,a) as before
-9 -7 -9 -7
’a;i = qa;i
Three Variables
l0
l1
![Page 77: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/77.jpg)
Va Vb
9
11 9-9
Vc
11 12-11
-9
-9
9
Forward Pass Backward Pass
-9 -7 -9 -7
All min-marginals are computed
Three Variables
l0
l1
![Page 78: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/78.jpg)
Belief Propagation on Chains
Start from left, go to right
Reparameterize current edge (a,b)
Mab;k = mini { a;i + ab;ik }
’ab;ik = ab;ik+ Mab;k - Mab;k’b;k = b;k
Repeat till the end of the chain
Start from right, go to left
Repeat till the end of the chain
![Page 79: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/79.jpg)
Belief Propagation on Chains
• A way of computing reparam constants
• Generalizes to chains of any length
• Forward Pass - Start to End
• MAP estimate
• Min-marginals of final variable
• Backward Pass - End to start
• All other min-marginals
Won’t need this .. But good to know
![Page 80: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/80.jpg)
Computational Complexity
• Each constant takes O(|L|)
• Number of constants - O(|E||L|)
O(|E||L|2)
• Memory required ?
O(|E||L|)
![Page 81: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/81.jpg)
Belief Propagation on Trees
Vb
Va
Forward Pass: Leaf Root
All min-marginals are computed
Backward Pass: Root Leaf
Vc
Vd Ve Vg Vh
![Page 82: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/82.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
– Exact MAP for Chains and Trees
– Approximate MAP for general graphs
– Computational Issues and Theoretical Properties
• Tree-reweighted Message Passing
![Page 83: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/83.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
Where do we start? Arbitrarily
a;0
a;1
b;0
b;1
d;0
d;1
c;0
c;1
Reparameterize (a,b)
![Page 84: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/84.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
a;0
a;1
’b;0
’b;1
d;0
d;1
c;0
c;1
Potentials along the red path add up to 0
![Page 85: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/85.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
a;0
a;1
’b;0
’b;1
d;0
d;1
’c;0
’c;1
Potentials along the red path add up to 0
![Page 86: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/86.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
a;0
a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Potentials along the red path add up to 0
![Page 87: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/87.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Potentials along the red path add up to 0
![Page 88: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/88.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Potentials along the red path add up to 0
- a;0
- a;1
’a;0 - a;0 = qa;0 ’a;1 - a;1 = qa;1
![Page 89: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/89.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Pick minimum min-marginal. Follow red path.
- a;0
- a;1
’a;0 - a;0 = qa;0 ’a;1 - a;1 = qa;1
![Page 90: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/90.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
a;0
a;1
’b;0
’b;1
d;0
d;1
c;0
c;1
Potentials along the red path add up to 0
![Page 91: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/91.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
a;0
a;1
’b;0
’b;1
d;0
d;1
’c;0
’c;1
Potentials along the red path add up to 0
![Page 92: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/92.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
a;0
a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Potentials along the red path add up to 0
![Page 93: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/93.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Potentials along the red path add up to 0
- a;0
- a;1
’a;1 - a;1 = qa;1 ’a;0 - a;0 ≤ qa;0≤
![Page 94: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/94.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Problem Solved
- a;0
- a;1
’a;1 - a;1 = qa;1 ’a;0 - a;0 ≤ qa;0≤
![Page 95: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/95.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
Problem Not Solved
- a;0
- a;1
’a;1 - a;1 = qa;1 ’a;0 - a;0 ≤ qa;0≥
![Page 96: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/96.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’b;0
’b;1
’d;0
’d;1
’c;0
’c;1
- a;0
- a;1
Reparameterize (a,b) again
![Page 97: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/97.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’’b;0
’’b;1
’d;0
’d;1
’c;0
’c;1
Reparameterize (a,b) again
But doesn’t this overcount some potentials?
![Page 98: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/98.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’’b;0
’’b;1
’d;0
’d;1
’c;0
’c;1
Reparameterize (a,b) again
Yes. But we will do it anyway
![Page 99: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/99.jpg)
Belief Propagation on Cycles
Va Vb
Vd Vc
’a;0
’a;1
’’b;0
’’b;1
’d;0
’d;1
’c;0
’c;1
Keep reparameterizing edges in some order
Hope for convergence and a good solution
![Page 100: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/100.jpg)
Belief Propagation
• Generalizes to any arbitrary random field
• Complexity per iteration ?
O(|E||L|2)
• Memory required ?
O(|E||L|)
![Page 101: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/101.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
– Exact MAP for Chains and Trees
– Approximate MAP for general graphs
– Computational Issues and Theoretical Properties
• Tree-reweighted Message Passing
![Page 102: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/102.jpg)
Computational Issues of BP
Complexity per iteration O(|E||L|2)
Special Pairwise Potentials ab;ik = wabd(|i-k|)
i - k
d
Potts
i - k
d
Truncated Linear
i - k
d
Truncated Quadratic
O(|E||L|) Felzenszwalb & Huttenlocher, 2004
![Page 103: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/103.jpg)
Computational Issues of BP
Memory requirements O(|E||L|)
Half of original BP Kolmogorov, 2006
Some approximations exist
But memory still remains an issue
Yu, Lin, Super and Tan, 2007
Lasserre, Kannan and Winn, 2007
![Page 104: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/104.jpg)
Computational Issues of BP
Order of reparameterization
Randomly
Residual Belief Propagation
In some fixed order
The one that results in maximum change
Elidan et al. , 2006
![Page 105: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/105.jpg)
Theoretical Properties of BP
Exact for Trees Pearl, 1988
What about any general random field?
Run BP. Assume it converges.
![Page 106: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/106.jpg)
Theoretical Properties of BP
Exact for Trees Pearl, 1988
What about any general random field?
Choose variables in a tree. Change their labels.
Value of energy does not decrease
![Page 107: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/107.jpg)
Theoretical Properties of BP
Exact for Trees Pearl, 1988
What about any general random field?
Choose variables in a cycle. Change their labels.
Value of energy does not decrease
![Page 108: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/108.jpg)
Theoretical Properties of BP
Exact for Trees Pearl, 1988
What about any general random field?
For cycles, if BP converges then exact MAP
Weiss and Freeman, 2001
![Page 109: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/109.jpg)
Results
Object Detection Felzenszwalb and Huttenlocher, 2004
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
H
TA1 A2
L1 L2
Labels - Poses of parts
Unary Potentials:
Fraction of foreground pixels
Pairwise Potentials:
Favour Valid Configurations
![Page 110: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/110.jpg)
Results
Object Detection Felzenszwalb and Huttenlocher, 2004
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
![Page 111: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/111.jpg)
Results
Binary Segmentation Szeliski et al. , 2008
Labels - {foreground, background}
Unary Potentials: -log(likelihood) using learnt fg/bg models
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
![Page 112: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/112.jpg)
Results
Binary Segmentation
Labels - {foreground, background}
Unary Potentials: -log(likelihood) using learnt fg/bg models
Szeliski et al. , 2008
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
Belief Propagation
![Page 113: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/113.jpg)
Results
Binary Segmentation
Labels - {foreground, background}
Unary Potentials: -log(likelihood) using learnt fg/bg models
Szeliski et al. , 2008
Global optimum
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
![Page 114: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/114.jpg)
ResultsSzeliski et al. , 2008
Labels - {disparities}
Unary Potentials: Similarity of pixel colours
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
Stereo Correspondence
![Page 115: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/115.jpg)
ResultsSzeliski et al. , 2008
Labels - {disparities}
Unary Potentials: Similarity of pixel colours
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
Belief Propagation
Stereo Correspondence
![Page 116: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/116.jpg)
ResultsSzeliski et al. , 2008
Labels - {disparities}
Unary Potentials: Similarity of pixel colours
Global optimum
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
Stereo Correspondence
![Page 117: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/117.jpg)
Summary of BP
Exact for chains
Exact for trees
Approximate MAP for general cases
Not even convergence guaranteed
So can we do something better?
![Page 118: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/118.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
– Integer Programming Formulation
– Linear Programming Relaxation and its Dual
– Convergent Solution for Dual
– Computational Issues and Theoretical Properties
![Page 119: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/119.jpg)
TRW Message Passing
• Convex (not Combinatorial) Optimization
• A different look at the same problem
• A similar solution
• Combinatorial (not Convex) Optimization
We will look at the most general MAP estimation
Not trees No assumption on potentials
![Page 120: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/120.jpg)
Things to Remember
• Forward-pass computes min-marginals of root
• BP is exact for trees
• Every iteration provides a reparameterization
• Basics of Mathematical Optimization
![Page 121: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/121.jpg)
Mathematical Optimization
min g0(x)
subject to gi(x) ≤ 0
i=1, … , N
• Objective function • Constraints
• Feasible region = {x | gi(x) ≤ 0}
x* = arg
Optimal
Solution
g0(x*)
Optimal
Value
![Page 122: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/122.jpg)
Integer Programming
min g0(x)
subject to gi(x) ≤ 0
i=1, … , N
• Objective function • Constraints
• Feasible region = {x | gi(x) ≤ 0}
x* = arg
Optimal
Solution
g0(x*)
Optimal
Value
xk Z
![Page 123: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/123.jpg)
Feasible Region
Generally NP-hard to optimize
![Page 124: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/124.jpg)
Linear Programming
min g0(x)
subject to gi(x) ≤ 0
i=1, … , N
• Objective function • Constraints
• Feasible region = {x | gi(x) ≤ 0}
x* = arg
Optimal
Solution
g0(x*)
Optimal
Value
![Page 125: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/125.jpg)
Linear Programming
min g0(x)
subject to gi(x) ≤ 0
i=1, … , N
• Linear objective function • Linear constraints
• Feasible region = {x | gi(x) ≤ 0}
x* = arg
Optimal
Solution
g0(x*)
Optimal
Value
![Page 126: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/126.jpg)
Linear Programming
min cTx
subject to Ax ≤ b
i=1, … , N
• Linear objective function • Linear constraints
• Feasible region = {x | Ax ≤ b}
x* = arg
Optimal
Solution
cTx*
Optimal
Value
Polynomial-time Solution
![Page 127: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/127.jpg)
Feasible Region
Polynomial-time Solution
![Page 128: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/128.jpg)
Feasible Region
Optimal solution lies on a vertex (obj func linear)
![Page 129: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/129.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
– Integer Programming Formulation
– Linear Programming Relaxation and its Dual
– Convergent Solution for Dual
– Computational Issues and Theoretical Properties
![Page 130: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/130.jpg)
Integer Programming Formulation
Va Vb
Label l0
Label l12
5
4
2
0
1 1
0
2
Unary Potentials
a;0 = 5
a;1 = 2
b;0 = 2
b;1 = 4
Labelling
f(a) = 1
f(b) = 0
ya;0 = 0 ya;1 = 1
yb;0 = 1 yb;1 = 0
Any f(.) has equivalent boolean variables ya;i
![Page 131: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/131.jpg)
Integer Programming Formulation
Va Vb
2
5
4
2
0
1 1
0
2
Unary Potentials
a;0 = 5
a;1 = 2
b;0 = 2
b;1 = 4
Labelling
f(a) = 1
f(b) = 0
ya;0 = 0 ya;1 = 1
yb;0 = 1 yb;1 = 0
Find the optimal variables ya;i
Label l0
Label l1
![Page 132: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/132.jpg)
Integer Programming Formulation
Va Vb
2
5
4
2
0
1 1
0
2
Unary Potentials
a;0 = 5
a;1 = 2
b;0 = 2
b;1 = 4
Sum of Unary Potentials
∑a ∑i a;i ya;i
ya;i {0,1}, for all Va, li
∑i ya;i = 1, for all Va
Label l0
Label l1
![Page 133: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/133.jpg)
Integer Programming Formulation
Va Vb
2
5
4
2
0
1 1
0
2
Pairwise Potentials
ab;00 = 0
ab;10 = 1
ab;01 = 1
ab;11 = 0
Sum of Pairwise Potentials
∑(a,b) ∑ik ab;ik ya;iyb;k
ya;i {0,1}
∑i ya;i = 1
Label l0
Label l1
![Page 134: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/134.jpg)
Integer Programming Formulation
Va Vb
2
5
4
2
0
1 1
0
2
Pairwise Potentials
ab;00 = 0
ab;10 = 1
ab;01 = 1
ab;11 = 0
Sum of Pairwise Potentials
∑(a,b) ∑ik ab;ik yab;ik
ya;i {0,1}
∑i ya;i = 1
yab;ik = ya;i yb;k
Label l0
Label l1
![Page 135: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/135.jpg)
Integer Programming Formulation
min ∑a ∑i a;i ya;i + ∑(a,b) ∑ik ab;ik yab;ik
ya;i {0,1}
∑i ya;i = 1
yab;ik = ya;i yb;k
![Page 136: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/136.jpg)
Integer Programming Formulation
min Ty
ya;i {0,1}
∑i ya;i = 1
yab;ik = ya;i yb;k
= [ … a;i …. ; … ab;ik ….]
y = [ … ya;i …. ; … yab;ik ….]
![Page 137: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/137.jpg)
One variable, two labels
ya;0
ya;1
ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1
y = [ ya;0 ya;1] = [ a;0 a;1]
![Page 138: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/138.jpg)
Two variables, two labels
= [ a;0 a;1 b;0 b;1
ab;00 ab;01 ab;10 ab;11]
y = [ ya;0 ya;1 yb;0 yb;1
yab;00 yab;01 yab;10 yab;11]
ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1
yb;0 {0,1} yb;1 {0,1} yb;0 + yb;1 = 1
yab;00 = ya;0 yb;0 yab;01 = ya;0 yb;1
yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1
![Page 139: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/139.jpg)
In General
Marginal Polytope
![Page 140: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/140.jpg)
In General
R(|V||L| + |E||L|2)
y {0,1}(|V||L| + |E||L|2)
Number of constraints
|V||L| + |V| + |E||L|2ya;i {0,1} ∑i ya;i = 1 yab;ik = ya;i yb;k
![Page 141: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/141.jpg)
Integer Programming Formulation
min Ty
ya;i {0,1}
∑i ya;i = 1
yab;ik = ya;i yb;k
= [ … a;i …. ; … ab;ik ….]
y = [ … ya;i …. ; … yab;ik ….]
![Page 142: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/142.jpg)
Integer Programming Formulation
min Ty
ya;i {0,1}
∑i ya;i = 1
yab;ik = ya;i yb;k
Solve to obtain MAP labelling y*
![Page 143: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/143.jpg)
Integer Programming Formulation
min Ty
ya;i {0,1}
∑i ya;i = 1
yab;ik = ya;i yb;k
But we can’t solve it in general
![Page 144: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/144.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
– Integer Programming Formulation
– Linear Programming Relaxation and its Dual
– Convergent Solution for Dual
– Computational Issues and Theoretical Properties
![Page 145: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/145.jpg)
Linear Programming Relaxation
min Ty
ya;i {0,1}
∑i ya;i = 1
yab;ik = ya;i yb;k
Two reasons why we can’t solve this
![Page 146: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/146.jpg)
Linear Programming Relaxation
min Ty
ya;i [0,1]
∑i ya;i = 1
yab;ik = ya;i yb;k
One reason why we can’t solve this
![Page 147: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/147.jpg)
Linear Programming Relaxation
min Ty
ya;i [0,1]
∑i ya;i = 1
∑k yab;ik = ∑kya;i yb;k
One reason why we can’t solve this
![Page 148: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/148.jpg)
Linear Programming Relaxation
min Ty
ya;i [0,1]
∑i ya;i = 1
One reason why we can’t solve this
= 1∑k yab;ik = ya;i∑k yb;k
![Page 149: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/149.jpg)
Linear Programming Relaxation
min Ty
ya;i [0,1]
∑i ya;i = 1
∑k yab;ik = ya;i
One reason why we can’t solve this
![Page 150: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/150.jpg)
Linear Programming Relaxation
min Ty
ya;i [0,1]
∑i ya;i = 1
∑k yab;ik = ya;i
No reason why we can’t solve this*
*memory requirements, time complexity
![Page 151: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/151.jpg)
One variable, two labels
ya;0
ya;1
ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1
y = [ ya;0 ya;1] = [ a;0 a;1]
![Page 152: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/152.jpg)
One variable, two labels
ya;0
ya;1
ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1
y = [ ya;0 ya;1] = [ a;0 a;1]
![Page 153: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/153.jpg)
Two variables, two labels
= [ a;0 a;1 b;0 b;1
ab;00 ab;01 ab;10 ab;11]
y = [ ya;0 ya;1 yb;0 yb;1
yab;00 yab;01 yab;10 yab;11]
ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1
yb;0 {0,1} yb;1 {0,1} yb;0 + yb;1 = 1
yab;00 = ya;0 yb;0 yab;01 = ya;0 yb;1
yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1
![Page 154: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/154.jpg)
Two variables, two labels
= [ a;0 a;1 b;0 b;1
ab;00 ab;01 ab;10 ab;11]
y = [ ya;0 ya;1 yb;0 yb;1
yab;00 yab;01 yab;10 yab;11]
ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1
yb;0 [0,1] yb;1 [0,1] yb;0 + yb;1 = 1
yab;00 = ya;0 yb;0 yab;01 = ya;0 yb;1
yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1
![Page 155: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/155.jpg)
Two variables, two labels
= [ a;0 a;1 b;0 b;1
ab;00 ab;01 ab;10 ab;11]
y = [ ya;0 ya;1 yb;0 yb;1
yab;00 yab;01 yab;10 yab;11]
ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1
yb;0 [0,1] yb;1 [0,1] yb;0 + yb;1 = 1
yab;00 + yab;01 = ya;0
yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1
![Page 156: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/156.jpg)
Two variables, two labels
= [ a;0 a;1 b;0 b;1
ab;00 ab;01 ab;10 ab;11]
y = [ ya;0 ya;1 yb;0 yb;1
yab;00 yab;01 yab;10 yab;11]
ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1
yb;0 [0,1] yb;1 [0,1] yb;0 + yb;1 = 1
yab;00 + yab;01 = ya;0
yab;10 + yab;11 = ya;1
![Page 157: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/157.jpg)
In General
Marginal Polytope
Local
Polytope
![Page 158: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/158.jpg)
In General
R(|V||L| + |E||L|2)
y [0,1](|V||L| + |E||L|2)
Number of constraints
|V||L| + |V| + |E||L|
![Page 159: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/159.jpg)
Linear Programming Relaxation
min Ty
ya;i [0,1]
∑i ya;i = 1
∑k yab;ik = ya;i
No reason why we can’t solve this
![Page 160: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/160.jpg)
Linear Programming Relaxation
Extensively studied
Optimization
Schlesinger, 1976
Koster, van Hoesel and Kolen, 1998
Theory
Chekuri et al, 2001 Archer et al, 2004
Machine Learning
Wainwright et al., 2001
![Page 161: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/161.jpg)
Linear Programming Relaxation
Many interesting Properties
• Global optimal MAP for trees
Wainwright et al., 2001
But we are interested in NP-hard cases
• Preserves solution for reparameterization
![Page 162: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/162.jpg)
Linear Programming Relaxation
• Large class of problems
• Metric Labelling
• Semi-metric Labelling
Many interesting Properties - Integrality Gap
Manokaran et al., 2008
• Most likely, provides best possible integrality gap
![Page 163: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/163.jpg)
Linear Programming Relaxation
• A computationally useful dual
Many interesting Properties - Dual
Optimal value of dual = Optimal value of primal
Easier-to-solve
![Page 164: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/164.jpg)
Dual of the LP RelaxationWainwright et al., 2001
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
min Ty
ya;i [0,1]
∑i ya;i = 1
∑k yab;ik = ya;i
![Page 165: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/165.jpg)
Dual of the LP RelaxationWainwright et al., 2001
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
1
2
3
4 5 6
1
2
3
4 5 6
ii =
i ≥ 0
![Page 166: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/166.jpg)
Dual of the LP RelaxationWainwright et al., 2001
1
2
3
4 5 6
q*(1)
ii =
q*(2)
q*(3)
q*(4) q*(5) q*(6)
i q*(i)
Dual of LP
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
i ≥ 0
max
![Page 167: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/167.jpg)
Dual of the LP RelaxationWainwright et al., 2001
1
2
3
4 5 6
q*(1)
ii
q*(2)
q*(3)
q*(4) q*(5) q*(6)
Dual of LP
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
i ≥ 0
i q*(i)max
![Page 168: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/168.jpg)
Dual of the LP RelaxationWainwright et al., 2001
ii
max i q*(i)
I can easily compute q*(i)
I can easily maintain reparam constraint
So can I easily solve the dual?
![Page 169: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/169.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
– Integer Programming Formulation
– Linear Programming Relaxation and its Dual
– Convergent Solution for Dual
– Computational Issues and Theoretical Properties
![Page 170: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/170.jpg)
TRW Message PassingKolmogorov, 2006
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
VaVb Vc
VdVe Vf
VgVh Vi
1
2
3
1
2
3
4 5 6
4 5 6
ii
i q*(i)
Pick a variable Va
![Page 171: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/171.jpg)
TRW Message PassingKolmogorov, 2006
ii
i q*(i)
Vc Vb Va
1c;0
1c;1
1b;0
1b;1
1a;0
1a;1
Va Vd Vg
4a;0
4a;1
4d;0
4d;1
4g;0
4g;1
![Page 172: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/172.jpg)
TRW Message PassingKolmogorov, 2006
11 + 44 + rest
1 q*(1) + 4 q*(4) + K
Vc Vb Va Va Vd Vg
Reparameterize to obtain min-marginals of Va
1c;0
1c;1
1b;0
1b;1
1a;0
1a;1
4a;0
4a;1
4d;0
4d;1
4g;0
4g;1
![Page 173: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/173.jpg)
TRW Message PassingKolmogorov, 2006
1’1 + 4’4 + rest
Vc Vb Va
’1c;0
’1c;1
’1b;0
’1b;1
’1a;0
’1a;1
Va Vd Vg
’4a;0
’4a;1
’4d;0
’4d;1
’4g;0
’4g;1
One pass of Belief Propagation
1 q*(’1) + 4 q*(’4) + K
![Page 174: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/174.jpg)
TRW Message PassingKolmogorov, 2006
1’1 + 4’4 + rest
Vc Vb Va Va Vd Vg
Remain the same
1 q*(’1) + 4 q*(’4) + K
’1c;0
’1c;1
’1b;0
’1b;1
’1a;0
’1a;1
’4a;0
’4a;1
’4d;0
’4d;1
’4g;0
’4g;1
![Page 175: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/175.jpg)
TRW Message PassingKolmogorov, 2006
1’1 + 4’4 + rest
1 min{’1a;0,’1a;1} + 4 min{’4a;0,’4a;1} + K
Vc Vb Va Va Vd Vg
’1c;0
’1c;1
’1b;0
’1b;1
’1a;0
’1a;1
’4a;0
’4a;1
’4d;0
’4d;1
’4g;0
’4g;1
![Page 176: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/176.jpg)
TRW Message PassingKolmogorov, 2006
1’1 + 4’4 + rest
Vc Vb Va Va Vd Vg
Compute weighted average of min-marginals of Va
’1c;0
’1c;1
’1b;0
’1b;1
’1a;0
’1a;1
’4a;0
’4a;1
’4d;0
’4d;1
’4g;0
’4g;1
1 min{’1a;0,’1a;1} + 4 min{’4a;0,’4a;1} + K
![Page 177: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/177.jpg)
TRW Message PassingKolmogorov, 2006
1’1 + 4’4 + rest
Vc Vb Va Va Vd Vg
’’a;0 = 1’1a;0+ 4’4a;0
1 + 4
’’a;1 = 1’1a;1+ 4’4a;1
1 + 4
’1c;0
’1c;1
’1b;0
’1b;1
’1a;0
’1a;1
’4a;0
’4a;1
’4d;0
’4d;1
’4g;0
’4g;1
1 min{’1a;0,’1a;1} + 4 min{’4a;0,’4a;1} + K
![Page 178: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/178.jpg)
TRW Message PassingKolmogorov, 2006
1’’1 + 4’’4 + rest
Vc Vb Va Va Vd Vg
’1c;0
’1c;1
’1b;0
’1b;1
’’a;0
’’a;1
’’a;0
’’a;1
’4d;0
’4d;1
’4g;0
’4g;1
1 min{’1a;0,’1a;1} + 4 min{’4a;0,’4a;1} + K
’’a;0 = 1’1a;0+ 4’4a;0
1 + 4
’’a;1 = 1’1a;1+ 4’4a;1
1 + 4
![Page 179: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/179.jpg)
TRW Message PassingKolmogorov, 2006
1’’1 + 4’’4 + rest
Vc Vb Va Va Vd Vg
’1c;0
’1c;1
’1b;0
’1b;1
’’a;0
’’a;1
’’a;0
’’a;1
’4d;0
’4d;1
’4g;0
’4g;1
1 min{’1a;0,’1a;1} + 4 min{’4a;0,’4a;1} + K
’’a;0 = 1’1a;0+ 4’4a;0
1 + 4
’’a;1 = 1’1a;1+ 4’4a;1
1 + 4
![Page 180: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/180.jpg)
TRW Message PassingKolmogorov, 2006
1’’1 + 4’’4 + rest
Vc Vb Va Va Vd Vg
1 min{’’a;0,’’a;1} + 4 min{’’a;0,’’a;1} + K
’1c;0
’1c;1
’1b;0
’1b;1
’’a;0
’’a;1
’’a;0
’’a;1
’4d;0
’4d;1
’4g;0
’4g;1
’’a;0 = 1’1a;0+ 4’4a;0
1 + 4
’’a;1 = 1’1a;1+ 4’4a;1
1 + 4
![Page 181: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/181.jpg)
TRW Message PassingKolmogorov, 2006
1’’1 + 4’’4 + rest
Vc Vb Va Va Vd Vg
(1 + 4) min{’’a;0, ’’a;1} + K
’1c;0
’1c;1
’1b;0
’1b;1
’’a;0
’’a;1
’’a;0
’’a;1
’4d;0
’4d;1
’4g;0
’4g;1
’’a;0 = 1’1a;0+ 4’4a;0
1 + 4
’’a;1 = 1’1a;1+ 4’4a;1
1 + 4
![Page 182: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/182.jpg)
TRW Message PassingKolmogorov, 2006
1’’1 + 4’’4 + rest
Vc Vb Va Va Vd Vg
(1 + 4) min{’’a;0, ’’a;1} + K
’1c;0
’1c;1
’1b;0
’1b;1
’’a;0
’’a;1
’’a;0
’’a;1
’4d;0
’4d;1
’4g;0
’4g;1
min {p1+p2, q1+q2} min {p1, q1} + min {p2, q2}≥
![Page 183: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/183.jpg)
TRW Message PassingKolmogorov, 2006
1’’1 + 4’’4 + rest
Vc Vb Va Va Vd Vg
Objective function increases or remains constant
’1c;0
’1c;1
’1b;0
’1b;1
’’a;0
’’a;1
’’a;0
’’a;1
’4d;0
’4d;1
’4g;0
’4g;1
(1 + 4) min{’’a;0, ’’a;1} + K
![Page 184: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/184.jpg)
TRW Message Passing
Initialize i. Take care of reparam constraint
Choose random variable Va
Compute min-marginals of Va for all trees
Node-average the min-marginals
REPEAT
Kolmogorov, 2006
Can also do edge-averaging
![Page 185: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/185.jpg)
Example 1
Va Vb
0
1 1
0
2
5
4
2l0
l1
Vb Vc
0
2 3
1
4
2
6
3
Vc Va
1
4 1
0
6
3
6
4
2 =1 3 =11 =1
5 6 7
Pick variable Va. Reparameterize.
![Page 186: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/186.jpg)
Example 1
Va Vb
-3
-2 -1
-2
5
7
4
2
Vb Vc
0
2 3
1
4
2
6
3
Vc Va
-3
1 -3
-3
6
3
10
7
2 =1 3 =11 =1
5 6 7
Average the min-marginals of Va
l0
l1
![Page 187: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/187.jpg)
Example 1
Va Vb
-3
-2 -1
-2
7.5
7
4
2
Vb Vc
0
2 3
1
4
2
6
3
Vc Va
-3
1 -3
-3
6
3
7.5
7
2 =1 3 =11 =1
7 6 7
Pick variable Vb. Reparameterize.
l0
l1
![Page 188: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/188.jpg)
Example 1
Va Vb
-7.5
-7 -5.5
-7
7.5
7
8.5
7
Vb Vc
-5
-3 -1
-3
9
6
6
3
Vc Va
-3
1 -3
-3
6
3
7.5
7
2 =1 3 =11 =1
7 6 7
Average the min-marginals of Vb
l0
l1
![Page 189: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/189.jpg)
Example 1
Va Vb
-7.5
-7 -5.5
-7
7.5
7
8.75
6.5
Vb Vc
-5
-3 -1
-3
8.75
6.5
6
3
Vc Va
-3
1 -3
-3
6
3
7.5
7
2 =1 3 =11 =1
6.5 6.5 7
Value of dual does not increase
l0
l1
![Page 190: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/190.jpg)
Example 1
Va Vb
-7.5
-7 -5.5
-7
7.5
7
8.75
6.5
Vb Vc
-5
-3 -1
-3
8.75
6.5
6
3
Vc Va
-3
1 -3
-3
6
3
7.5
7
2 =1 3 =11 =1
6.5 6.5 7
Maybe it will increase for Vc
NO
l0
l1
![Page 191: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/191.jpg)
Example 1
Va Vb
-7.5
-7 -5.5
-7
7.5
7
8.75
6.5
Vb Vc
-5
-3 -1
-3
8.75
6.5
6
3
Vc Va
-3
1 -3
-3
6
3
7.5
7
2 =1 3 =11 =1
Strong Tree Agreement
Exact MAP Estimate
f1(a) = 0 f1(b) = 0 f2(b) = 0 f2(c) = 0 f3(c) = 0 f3(a) = 0
l0
l1
![Page 192: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/192.jpg)
Example 2
Va Vb
0
1 1
0
2
5
2
2
Vb Vc
1
0 0
1
0
0
0
0
Vc Va
0
1 1
0
0
3
4
8
2 =1 3 =11 =1
4 0 4
Pick variable Va. Reparameterize.
l0
l1
![Page 193: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/193.jpg)
Example 2
Va Vb
-2
-1 -1
-2
4
7
2
2
Vb Vc
1
0 0
1
0
0
0
0
Vc Va
0
0 1
-1
0
3
4
9
2 =1 3 =11 =1
4 0 4
Average the min-marginals of Va
l0
l1
![Page 194: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/194.jpg)
Example 2
Va Vb
-2
-1 -1
-2
4
8
2
2
Vb Vc
1
0 0
1
0
0
0
0
Vc Va
0
0 1
-1
0
3
4
8
2 =1 3 =11 =1
4 0 4
Value of dual does not increase
l0
l1
![Page 195: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/195.jpg)
Example 2
Va Vb
-2
-1 -1
-2
4
8
2
2
Vb Vc
1
0 0
1
0
0
0
0
Vc Va
0
0 1
-1
0
3
4
8
2 =1 3 =11 =1
4 0 4
Maybe it will decrease for Vb or Vc
NO
l0
l1
![Page 196: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/196.jpg)
Example 2
Va Vb
-2
-1 -1
-2
4
8
2
2
Vb Vc
1
0 0
1
0
0
0
0
Vc Va
0
0 1
-1
0
3
4
8
2 =1 3 =11 =1
f1(a) = 1 f1(b) = 1 f2(b) = 1 f2(c) = 0 f3(c) = 1 f3(a) = 1
f2(b) = 0 f2(c) = 1
Weak Tree Agreement
Not Exact MAP Estimate
l0
l1
![Page 197: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/197.jpg)
Example 2
Va Vb
-2
-1 -1
-2
4
8
2
2
Vb Vc
1
0 0
1
0
0
0
0
Vc Va
0
0 1
-1
0
3
4
8
2 =1 3 =11 =1
Weak Tree Agreement
Convergence point of TRW
l0
l1
f1(a) = 1 f1(b) = 1 f2(b) = 1 f2(c) = 0 f3(c) = 1 f3(a) = 1
f2(b) = 0 f2(c) = 1
![Page 198: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/198.jpg)
Obtaining the Labelling
Only solves the dual. Primal solutions?
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
’ = ii
Fix the label
Of Va
![Page 199: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/199.jpg)
Obtaining the Labelling
Only solves the dual. Primal solutions?
Va Vb Vc
Vd Ve Vf
Vg Vh Vi
’ = ii
Fix the label
Of Vb
Continue in some fixed order
Meltzer et al., 2006
![Page 200: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/200.jpg)
Outline
• Problem Formulation
• Reparameterization
• Belief Propagation
• Tree-reweighted Message Passing
– Integer Programming Formulation
– Linear Programming Relaxation and its Dual
– Convergent Solution for Dual
– Computational Issues and Theoretical Properties
![Page 201: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/201.jpg)
Computational Issues of TRW
• Speed-ups for some pairwise potentials
Basic Component is Belief Propagation
Felzenszwalb & Huttenlocher, 2004
• Memory requirements cut down by half
Kolmogorov, 2006
• Further speed-ups using monotonic chains
Kolmogorov, 2006
![Page 202: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/202.jpg)
Theoretical Properties of TRW
• Always converges, unlike BP
Kolmogorov, 2006
• Strong tree agreement implies exact MAP
Wainwright et al., 2001
• Optimal MAP for two-label submodular problems
Kolmogorov and Wainwright, 2005
ab;00 + ab;11 ≤ ab;01 + ab;10
![Page 203: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/203.jpg)
Results
Binary Segmentation Szeliski et al. , 2008
Labels - {foreground, background}
Unary Potentials: -log(likelihood) using learnt fg/bg models
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
![Page 204: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/204.jpg)
Results
Binary Segmentation
Labels - {foreground, background}
Unary Potentials: -log(likelihood) using learnt fg/bg models
Szeliski et al. , 2008
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
TRW
![Page 205: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/205.jpg)
Results
Binary Segmentation
Labels - {foreground, background}
Unary Potentials: -log(likelihood) using learnt fg/bg models
Szeliski et al. , 2008
Belief Propagation
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
![Page 206: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/206.jpg)
Results
Stereo Correspondence Szeliski et al. , 2008
Labels - {disparities}
Unary Potentials: Similarity of pixel colours
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
![Page 207: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/207.jpg)
ResultsSzeliski et al. , 2008
Labels - {disparities}
Unary Potentials: Similarity of pixel colours
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
TRW
Stereo Correspondence
![Page 208: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/208.jpg)
ResultsSzeliski et al. , 2008
Labels - {disparities}
Unary Potentials: Similarity of pixel colours
Belief Propagation
Pairwise Potentials: 0, if same labels
1 - exp(|Da - Db|), if different labels
Stereo Correspondence
![Page 209: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/209.jpg)
Results
Non-submodular problems Kolmogorov, 2006
BP TRW-S
30x30 grid K50
BP TRW-S
BP outperforms TRW-S
![Page 210: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/210.jpg)
Summary
• Trees can be solved exactly - BP
• No guarantee of convergence otherwise - BP
• Strong Tree Agreement - TRW-S
• Submodular energies solved exactly - TRW-S
• TRW-S solves an LP relaxation of MAP estimation
• Loopier graphs give worse results
Rother and Kolmogorov, 2006
![Page 211: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/211.jpg)
Related New(er) Work
• Solving the Dual
Globerson and Jaakkola, 2007
Komodakis, Paragios and Tziritas 2007
Weiss et al., 2006
Schlesinger and Giginyak, 2007
• Solving the Primal
Ravikumar, Agarwal and Wainwright, 2008
![Page 212: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/212.jpg)
Related New(er) Work
• More complex relaxations
Sontag and Jaakkola, 2007
Komodakis and Paragios, 2008
Kumar, Kolmogorov and Torr, 2007
Werner, 2008
Sontag et al., 2008
Kumar and Torr, 2008
![Page 213: PowerPoint Presentation - MAP Estimation Algorithms in ... · MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision](https://reader034.fdocuments.in/reader034/viewer/2022042217/5ebfeaa8681084319d2d821d/html5/thumbnails/213.jpg)
Questions on Part I ?
Code + Standard Data
http://vision.middlebury.edu/MRF