Publications

Highlights

Please also see Google Scholar.

For associated solvers and datasets see here.

Exact MAP-inference by Confining Combinatorial Search with LP Relaxation

We consider a new relaxation of the MAP-inference problem for MRFs that decomposes the problem into two parts: an easy LP-tight part and a difficult one. These parts are used by efficient message passing and exact ILP solvers respectively, resulting in reduced overall runtime.

Stefan Haller, Paul Swoboda, Bogdan Savchynskyy

AAAI 2018

A Dual Ascent Framework for Lagrangean Decomposition of Combinatorial Problems

We introduce a general framework for optimizing Lagrangean decompositions of combinatorial problems based on dual block coordinate ascent.

Paul Swoboda, Jan Kuske and Bogdan Savchynskyy

CVPR 2017

A study of Lagrangean decompositions and dual ascent solvers for graph matching

We propose a family of dual block coordinate ascent algorithms for the graph matching problem that surpass state of the art solvers.

Paul Swoboda, Carsten Rother, Hassan Abu Alhaija, Dagmar Kainmüller, Bogdan Savchynskyy

CVPR 2017

A Message Passing Algorithm for the Minimum Cost Multicut Problem

We propose a new message passing (a.k.a. dual block coordinate ascent) for the multicut problem (a.k.a. correlation clustering) that outperforms existing LP-based solvers and improves solutions of primal heuristics.

Paul Swoboda, Bjoern Andres

CVPR 2017

A Novel Convex Relaxation for Non-binary Discrete Tomography

We propose a new and tighter relaxation for the multilabel discrete tomography problem and algorithms for solving it.

Jan Kuske, Paul Swoboda, Stefania Petra

SSVM 2017

Maximum Persistency via Iterative Relaxed Inference with Graphical Models

We propose a new approach to partial optimality that surpasses previous methods both in terms of attained number of persistent variables and runtime.

Alexander Shekhovtsov, Paul Swoboda, Bogdan Savchynskyy

CVPR 2015

 

Full List

MAP inference via Block-Coordinate Frank-Wolfe Algorithm
Paul Swoboda, Vladimir Kolmogorov
arXiv

Exact MAP-inference by Confining Combinatorial Search with LP Relaxation
Stefan Haller, Paul Swoboda, Bogdan Savchynskyy
AAAI 2018

A Dual Ascent Framework for Lagrangean Decomposition of Combinatorial Problems
Paul Swoboda, Jan Kuske and Bogdan Savchynskyy
CVPR 2017

A study of Lagrangean decompositions and dual ascent solvers for graph matching
Paul Swoboda, Carsten Rother, Hassan Abu Alhaija, Dagmar Kainmüller, Bogdan Savchynskyy
CVPR 2017

A Message Passing Algorithm for the Minimum Cost Multicut Problem
Paul Swoboda, Bjoern Andres
CVPR 2017

A Novel Convex Relaxation for Non-binary Discrete Tomography
Jan Kuske, Paul Swoboda, Stefania Petra
SSVM 2017

Multicuts and Perturb & MAP for Probabilistic Graph Clustering
Jörg Hendrik Kappes, Paul Swoboda, Bogdan Savchynskyy, Tamir Hazan, Christoph Schnörr
Journal of Mathematical Imaging and Vision, 2016

Partial Optimality by Pruning for MAP-Inference with General Graphical Models
Paul Swoboda, Alexander Shekhovtsov, Jörg Hendrik Kappes, Christoph Schnörr, Bogdan Savchynskyy
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016

Probabilistic correlation clustering and image partitioning using perturbed multicuts
Jörg Hendrik Kappes, Paul Swoboda, Bogdan Savchynskyy, Tamir Hazan, Christoph Schnörr
SSVM 2015

Maximum Persistency via Iterative Relaxed Inference with Graphical Models
Alexander Shekhovtsov, Paul Swoboda, Bogdan Savchynskyy
CVPR 2015

Convex variational image restoration with histogram priors
Paul Swoboda, Christoph Schnörr
SIAM Journal on Imaging Sciences 2013

Variational image segmentation and cosegmentation with the wasserstein distance
Paul Swoboda, Christoph Schnörr
EMMCVPR 2013

Partial optimality via iterative pruning for the Potts model
Paul Swoboda, Bogdan Savchynskyy, Jörg Kappes, Christoph Schnörr
SSVM 2013

Global MAP-optimality by shrinking the combinatorial search area with convex relaxation
Bogdan Savchynskyy, Jörg Hendrik Kappes, Paul Swoboda, Christoph Schnörr
NIPS 2013