Invited Session Mon.1.H 2053

Monday, 10:30 - 12:00 h, Room: H 2053

Cluster 9: Global optimization [...]

Optimization models and methods for computer vision


Chair: Jiming Peng and Vikas Singh



Monday, 10:30 - 10:55 h, Room: H 2053, Talk 1

Vladimir Kolmogorov
Message passing algorithms for MAP-MRF inference


I will consider the problem of computing maximum a posterior configuration in a Markov Random Field, or equivalently minimizing a function of discrete variables decomposed as a sum of low-order terms. This task frequently occurs in many fields such as computer vision and machine learning. A popular approach to tackling this NP-hard problem is to solve its LP relaxation. I will talk about message passing algorithms that try to solve the LP, in particular sequential tree-reweighted message passing (TRW-S) and its extensions. TRW-S shows good performance in practice and is often used for computer vision problems.



Monday, 11:00 - 11:25 h, Room: H 2053, Talk 2

Daniel Cremers
Convex relaxation techniques with applications in computer vision

Coauthors: Antonin Chambolle, Bastian Goldl├╝cke, Kalin Kolev, Thomas Pock, Evgeny Strekalovksiy


Numerous computer vision problems can be solved by variational methods
and partial differential equations. Yet, many traditional approaches
correspond to non-convex energies giving rise to suboptimal solutions
and often strong dependency on appropriate initialization. In my
presentation, I will show how problems like image segmentation,
multiview stereo reconstruction and optic flow estimation can be
formulated as variational problems. Subsequently, I will introduce
convex relaxation techniques which allow to compute globally optimal or
near-optimal solutions. The resulting algorithms provide robust
solutions, independent of initialization and compare favorable to
spatially discrete graph theoretic approaches in terms of computation
time, memory requirements and accuracy.



Monday, 11:30 - 11:55 h, Room: H 2053, Talk 3

Maxwell D. Collins
Random walks based multi-image segmentation: Quasiconvexity results and GPU-based solutions

Coauthors: Leo Grady, Vikas Singh, Jia Xu


We recast the cosegmentation problem using random Walker segmentation as the core segmentation algorithm, rather than the traditional MRF approach adopted in the literature so far. Our formulation is similar to previous approaches in that it also permits cosegmentation constraints which impose consistency between the extracted objects from 2+ images using a nonparametric model. However, several previous nonparametric cosegmentation methods have the limitation that they require one auxiliary node for every pair of pixels that are similar (limiting such methods to describing only those objects that have high entropy appearance models). Our proposed model eliminates this dependence - the resulting improvements are significant. We further allow an optimization scheme exploiting quasiconvexity for model-based segmentation with no dependence on the scale of the segmented foreground. Finally, we show that the optimization can be expressed in terms of operations on sparse matrices which are easily mapped to GPU architecture. We provide a specialized CUDA library for cosegmentation exploiting this special structure, and report experimental results showing these advantages.


  personal loans. But at the same time, it acts only with sexual arousal. Viagra has a number of advantages in comparison with injections in the sexual organ or other procedures aimed at treatment of impotency.