43
Extensions of submodularity and their application in computer vision Vladimir Kolmogorov IST Austria Oxford, 20 January 2014

Extensions of submodularity and their application in

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Extensions of submodularity and their application in

Extensions of submodularity and

their application in computer vision

Vladimir Kolmogorov

IST Austria

Oxford, 20 January 2014

Page 2: Extensions of submodularity and their application in

Linear Programming relaxation

• Popular approach: Basic LP relaxation (BLP)

- for pairwise energies: Schlesinger’s LP

- for higher-order energies: enforce consistency between variables and for all and

- many algorithms for (approximately) solving it (MSD, TRW-S, MPLP, ...)

Page 3: Extensions of submodularity and their application in

Tightness of BLP

Theorem [Cooper’08], [Werner’10].

If each term is a submodular function

then BLP relaxation is tight.

• Other such classes? Complete classification?

• Use VCSP framework

Page 4: Extensions of submodularity and their application in

Valued Constraint Satisfaction Problem (VCSP)

• Language G: a set of cost functions

• VCSP(G): class of functions that can be expressed as a sum of functions from G with overlapping sets of vars

- Goal: minimize this sum

• Complexity of G?

• Does BLP solves G?

• Feder-Vardi conjecture (for CSPs):

Every CSP language is either tractable or NP-hard

- CSP: contains functions

Page 5: Extensions of submodularity and their application in

Classifications for finite-valued CSPs

- Other languages are NP-hard [Thapper,Živný STOC’13]

Theorem [Thapper,Živný FOCS’12], [K ICALP’13]

BLP solves G iff it admits a binary symmetric fractional polymorphism

Page 6: Extensions of submodularity and their application in

Submodular functions

0

1

2

3

4

Page 7: Extensions of submodularity and their application in

New classes of functions

Page 8: Extensions of submodularity and their application in

Useful classes

• Submodular functions

- Pairwise functions: can be solved via maxflow (“graph cuts”)

- Lots of applications in computer vision

• Bisubmodular functions

- Obtaining partial optimal solutions

- Characterize extensions of “QPBO” to arbitrary pseudo-Boolean functions [K’10,12]

• k-submodular functions

- Partial optimality for functions of k-valued variables

- This talk: efficient algorithm for Potts energy [Gridchyn, K ICCV’13]

Page 9: Extensions of submodularity and their application in

Partial optimality

• Input: function

• Partial labeling is optimal if it can be extended

to a full optimal labeling

Page 10: Extensions of submodularity and their application in

Partial optimality

• Input: function

• Partial labeling is optimal if it can be extended

to a full optimal labeling

• Can be viewed as a labeling

Page 11: Extensions of submodularity and their application in

k-submodular relaxations

• Input: function

Page 12: Extensions of submodularity and their application in

k-submodular relaxations

• Input: function

• Construct extension

which is k-submodular

• Minimize

Theorem:

Minimum of partially optimal

Page 13: Extensions of submodularity and their application in

k-submodularity

• Function is k-submodular if

Page 14: Extensions of submodularity and their application in

k-submodular relaxations

• Case k = 2 ([K’10,12])

Bisubmodular relaxation

Characterizes extensions of QPBO

• Case k > 2

[Gridchyn,K ICCV’13] :

- efficient method for Potts energies

[Wahlström SODA’14] :

- used for FPT algorithms

Page 15: Extensions of submodularity and their application in

k-submodular relaxations for Potts energy

• k-submodular relaxation:

Page 16: Extensions of submodularity and their application in

k-submodular relaxations for Potts energy

• k-submodular relaxation:

: tree metric

Page 17: Extensions of submodularity and their application in

k-submodular relaxations for Potts energy

• k-submodular relaxation:

(·) : k-submodular relaxation of (·)

3 4 0 10

1.5

Page 18: Extensions of submodularity and their application in

k-submodular relaxations for Potts energy

• Minimizing : O(log k) maxflows

• Alternative approach: [Kovtun ’03,’04]

- Stronger than k-submodular relaxations (labels more)

- Can be solved by the same approach!

complexity: k => O(log k) maxflows

- Part of “Reduce, Reuse, Recycle’’ [Alahari et al.’08,’10]

- Our tests for stereo: 50-93% labeled

with 9x9 windows

- Speeds up alpha-expansion for unlabeled part

Page 19: Extensions of submodularity and their application in

Tree Metrics

• [Felzenszwalb et al.’10]: O(log k) maxflows for

Page 20: Extensions of submodularity and their application in

Tree Metrics

• [Felzenszwalb et al.’10]: O(log k) maxflows for

• [This work]: extension to more general unary terms

- new proof of correctness

Page 21: Extensions of submodularity and their application in

Special case: Total Variation

• Convex unary terms

• Reduction to parametric maxflow [Hochbaum’01],

[Chambolle’05], [Darbon,Sigelle’05]

Page 22: Extensions of submodularity and their application in

New condition: T-convexity

• Convexity for any pair of adjacent edges:

Page 23: Extensions of submodularity and their application in

Algorithm: divide-and-conquer

• Pick edge ( )

• Compute

Page 24: Extensions of submodularity and their application in

Algorithm: divide-and-conquer

• Pick edge ( )

• Compute

• Claim: has a minimizer as shown below

• Solve two subproblems recursively

Page 25: Extensions of submodularity and their application in

Achieving balanced splits

• For star graphs, all splits are unbalanced

• Solution [Felzenszwalb et al.’10]: insert a new short edge

- modify unary terms (·) accordingly

Page 26: Extensions of submodularity and their application in

Algorithm illustration

• k = 7 labels:

1,2,3,4,5,6,7

1

2 3

4

5

6

7

5

6

7

1

2 3

4

Page 27: Extensions of submodularity and their application in

Algorithm illustration

• k = 7 labels:

1,2,3,4

5,6,7

1

2 3

4

5

6

7

5

6

7

1

2 3

4

Page 28: Extensions of submodularity and their application in

Algorithm illustration

• k = 7 labels:

5,6,7 1,2

3,4

Page 29: Extensions of submodularity and their application in

Algorithm illustration

• k = 7 labels:

1,2

3,4

5,6

7

Page 30: Extensions of submodularity and their application in

Algorithm illustration

• k = 7 labels:

7

1

2

3 4

5

6

“Kovtun labeling” unlabeled part,

run alpha-expansion

• maxflows

Page 31: Extensions of submodularity and their application in

Stereo results

Page 32: Extensions of submodularity and their application in

Proof of correctness (sketch)

Page 33: Extensions of submodularity and their application in

Proof of correctness (sketch)

• For labeling and edge ( ) define

Page 34: Extensions of submodularity and their application in

Proof of correctness (sketch)

• For labeling and edge ( ) define

Page 35: Extensions of submodularity and their application in

Proof of correctness (sketch)

• Coarea formula:

Page 36: Extensions of submodularity and their application in

Proof of correctness (sketch)

• Coarea formula:

Page 37: Extensions of submodularity and their application in

Proof of correctness (sketch)

• Coarea formula:

Page 38: Extensions of submodularity and their application in

Proof of correctness (sketch)

• Coarea formula:

Page 39: Extensions of submodularity and their application in

Proof of correctness (sketch)

• Coarea formula:

Page 40: Extensions of submodularity and their application in

Proof of correctness (sketch)

• Coarea formula:

• Equivalent problem: minimize

with subject to consistency constraints

• Equivalent to independent minimizations of

- consistency holds automatically due to convexity of

Page 41: Extensions of submodularity and their application in

Extension to trees

• Coarea formula:

Page 42: Extensions of submodularity and their application in

Summary

Part I:

• New tractable class of functions

complete characterization for finite-valued CSPs

Part II:

• k-submodular relaxations for partial optimality

• For Potts model:

cast Kovtun’s approach as k-submodular function minimization

O(log k) algorithm

generalized alg. of [Felzenswalb et al’10] for tree metrics

• Future work: k-submodular relaxations for other functions?

Page 43: Extensions of submodularity and their application in

postdoc & PhD student

positions are available