Keynote Speakers

• Stephen Boyd  (Stanford University)

Distributed Optimization via the Alternating Direction Method of Multipliers

Joint work with Neal Parikh, Eric Chu, Borja Peleato, and Jon Eckstein

Abstract

Problems in areas such as machine learning and dynamic optimization on a large network lead to extremely large convex optimization problems, with problem data stored in a decentralized way, and processing elements
distributed across a network. We argue that the alternating direction method of multipliers is well suited to such problems.  The method was developed in the 1970s, with roots in the 1950s, and is equivalent or closely related to many other algorithms, such as dual decomposition, the method of multipliers, Douglas-Rachford splitting, Spingarn’s method of partial inverses, Dykstra’s alternating projections, Bregman iterative algorithms for $\ell_1$ problems, proximal methods, and others. After briefly surveying the theory and history of the algorithm, we discuss applications to statistical and machine learning problems such as the lasso and support vector machines, and to dynamic energy management problems arising in the smart grid.

• Olga Veksler (University of Western Ontario)

Pixel Labeling Problems With Structured Layout

Abstract

Pixel labeling problems are pervasive in computer vision. In this talk, we discuss optimization approaches for  problems where we have some structure imposed on the layout of the labels. In other words, the relationships between labels is not arbitrary, but has a well defined spatial structure. Imposing a structure on label layout is useful for applications such as estimating rough geometric surface orientation of a scene or for imposing various shape priors on an object to be segmented from an image. For example, we may want to have pixels labeled as a “left” facing surface to be to the left of pixels labeled as a “right” facing surface.  We first show that popular energy minimization algorithms, such as the expansion algorithm, easily get stuck in a local minimum for such structured layout problems. Then we  describe various approaches for structured layout scenes. These approaches are  based either on dynamic programming or graph-cut move-making algorithms. In some interesting cases, a global minimum is guaranteed, while in others, a strong local minimum with respect to a certain move space is found. We show applications to geometric class labeling and segmentation with a shape prior.

• Thomas Pock (Graz University of Technology)

Convex Relaxation of Curvature Related Functionals

Joint work with Kristian Bredies, University Graz, Benedikt Wirth, Courant Institute of Mathematical Sciences

Abstract

We investigate a class of variational problems that incorporate curvature information of the level lines. We consider functionals incorporating metrics defined on the orientations of pairs of line segments that meet in the vertices of the level lines. We discuss two particular instances: One instance that minimizes the total number of vertices of the level lines and another instance that minimizes the total sum of the absolute exterior angles between the line segments. In case of smooth level lines, the latter corresponds to the total absolute curvature. We show that these problems can be solved approximately by means of a tractable convex relaxation in higher dimensions. Furthermore, we will propose a generalization of the framework to the functionals depending on the squared curvature and show that the total variation plays a crucial role in the relaxation of such curvature incorporating functionals. In our numerical experiments we present results for image segmentation, image denoising and image inpainting.

EMMCVPR 2013 Final Program

August 19-21, 2013

 Monday 08:15-08:30 Welcome, Practical Information (Fredrik Kahl) 08:30-09:20 Session 1: Keynote Speaker (Session Chair: Fredrik Kahl) Distributed Optimization via the Alternating Direction Method of Multipliers Stephen Boyd Stanford University 09:20-10:10 Session 2: Medical Imaging (Session Chair: Niels Christian Overgaard) Rapid Mode Estimation for 3D Brain MRI Tumor Segmentation Haithem Boussaid, Iasonas Kokkinos, Nikos Paragios Ecole Centrale Paris Jointly Segmenting Prostate Zones in 3D MRIs by Globally Optimized Coupled Level-Sets Jing Yuan, Eranga Ukwatta, Wu Qiu, Martin Rajchl, Yue Sun, Xue-Cheng Tai, Aaron Fenster University of Western Ontario, University of Bergen 10:10-10:40 Coffee 10:40-12:00 Session 3: Image Editing (Session Chair: Xue-Cheng Tai) Linear Osmosis Models for Visual Computing Joachim Weickert, Kai Hagenburg, Michael Breuss, Oliver Vogel Saarland University, Brandenburg University of Technology Analysis of Bayesian Blind Deconvolution David Wipf, Haichao Zhang Microsoft Research, Northwestern Polytechnical University A Variational Method for Expanding the Bit-Depth of Low Contrast Image Motong Qiao, Wei Wang, and Michael K. Ng Hong Kong Baptist University 12:00-13:20 Lunch 13:20-14:40 Session 4: 3D Reconstruction (Session Chair: Kalle Åström) Variational Shape from Light Field Stefan Heber, Rene Ranftl, Thomas Pock TU Graz Simultaneous Fusion Moves for 3D-Label Stereo Johannes Ulén, Carl Olsson Lund University Efficient Convex Optimization for Minimal Partition Problems with Volume Constraints Thomas Möllenhoff, Claudia Nieuwenhuis, Eno Töppe, Daniel Cremers TU Munich 14:40-15:10 Coffee 15:10-16:30 Session 5:  Shape Matching (Session Chair: Andrew Fitzgibbon) Discrete Geodesic Regression in Shape Space Benjamin Berkels, Thomas Fletcher, Behrend Heeren, Martin Rumpf, Benedikt Wirth Universität Bonn, University of Utah, New York University Object Segmentation by Shape Matching with Wasserstein Modes Bernhard Schmitzer, Christoph Schnörr University of Heidelberg Learning a Model for Shape-Constrained Image Segmentation from Weakly Labeled Data Boris Yangel, Dmitry Vetrov Moscow State University Tuesday 08:30-09:20 Session 6: Keynote Speaker (Session Chair: Anders Heyden) Pixel Labeling Problems With Structured Layout Olga Veksler University of Western Ontario 09:20-10:10 Session 7: Image Restoration (Session Chair: Rudolf Mester) An Optimal Control Approach to Find Sparse Data for Laplace Interpolation Laurent Hoeltgen, Simon Setzer, Joachim Weickert Saarland University Curvature Regularization for Resolution-Independent Images John MacCormick, Andrew Fitzgibbon Dickinson College, Microsoft Research 10:10-10:40 Coffee 10:40-12:00 Session 8: Scene Understanding (Session Chair: Jan-Erik Solem) PoseField: An Efficient Mean-field based Method for Joint Estimation of Human Pose, … Segmentation, and Depth Vibhav Vineet, Glenn Sheasby, Jonathan Warrell, Philip H.S. Torr Oxford Brookes University Semantic Video Segmentation From Occlusion Relations Within a Convex Optimization Framework Brian Taylor, Alper Ayvaci, Avinash Ravichandran, Stefano Soatto UCLA, Honda Research Institute A Co-occurrence Prior for Continuous Multi-Label Optimization Mohamed Souiai, Evgeny Strekalovskiy, Claudia Nieuwenhuis, Daniel Cremers TU Munich 12:00-13:20 Lunch 13:20-14:40 Session 9: Segmentation I (Session Chair: Thomas Pock) Convex Relaxations for a Generalized Chan-Vese Model Egil Bae, Jan Lellman, Xue-Cheng Tai UCLA, Cambridge University, University of Bergen Multiclass Segmentation by Iterated ROF Thresholding Xiaohao Cai, Gabriele Steidl University of Kaiserslautern A Generic Convexification and Graph Cut Method for Multiphase Image Segmentation Jun Liu, Xue-Cheng Tai, Shingyu Leung Beijing Normal University, University of Bergen, Hong Kong University of Science and Technology 14:40-15:10 Coffee 15:10-16:00 Session 10: Superpixels (Session Chair: Olga Veksler) Segmenting Planar Superpixel Adjacency Graphs w.r.t. Non-planar Superpixel Affinity Graphs Bjoern Andres, Julian Yarkony, B. Manjunath, Steffen Kirchhoff, Engin Turetken, C. Fowlkes, H. Pfister Harvard University, UCSB, EPFL, UCI Contour-relaxed Superpixels Christian Conrad, Matthias Mertz, Rudolf Mester Goethe University Frankfurt, Linköping University Wednesday 08:30-09:20 Session 11: Keynote Speaker (Session Chair: Magnus Oskarsson) Convex Relaxation of Curvature Related Functionals Thomas Pock Graz University of Technology 09:20-10:10 Session 12: Statistical Methods and Learning (Session Chair: Cristian Sminchisescu) Sparse-MIML: A Sparsity-Based Multi-Instance Multi-Learning Algorithm Chenyang Shen, Michael Ng, LiPing Jing Hong Kong Baptist University, Beijing Jiaotong University Consensus Clustering with Robust Evidence Accumulation Andrè Lourenco, Samuel Rota Bulò, Ana Fred, Marcello Pelillo Instituto Superior de Engenharia de Lisboa, Instituto Superior Técnico Lisboa, Università Ca’ Foscari Venezia 10:10-10:40 Coffee 10:40-12:00 Session 13: Segmentation II (Session Chair: Joachim Weickert) Variational Image Segmentation and Cosegmentation with the Wasserstein Distance Paul Swoboda, Christoph Schnörr University of Heidelberg A Convex Formulation for Global Histogram Based Binary Segmentation Romain Yildizoglu, Jean-Francois Aujol, Nicolas Papadakis Université de Bordeaux A Continuous Shape Prior for MRF-based Segmentation Dmitrij Schlesinger TU Dresden