E-Mail: paul.swoboda (at) ist.ac.at
A Dual Ascent Framework for Lagrangean Decomposition of Combinatorial Problems
Paul Swoboda, Jan Kuske and Bogdan Savchynskyy, CVPR 2017. project page.
Study of Lagrangean Decomposition and Dual Ascent Solvers for Graph Matching
Paul Swoboda, Carsten Rother, Hassan Abu Alhaija, Dagmar Kainmüller and Bogdan Savchynskyy, CVPR 2017. code.
A Message Passing Algorithm for the Minimum Cost Multicut Problem
Paul Swoboda and Bjoern Andres, CVPR 2017. code.
Maximum Persistency via Iterative Relaxed Inference with Graphical Models
Alexander Shekhovtsov, Paul Swoboda, Bogdan Savchynskyy, CVPR 2015. pdf bib
project page: Persistency via Iterative Relaxed Inference.
Partial Optimality by Pruning for MAP-inference with General Graphical Models
Paul Swoboda, Bogdan Savchynskyy, Jörg H. Kappes and Christoph Schnörr, CVPR 2014. pdf bib
Accompanying code: Persistency by Pruning.
LP_MP: a C++ framework for convergent message passing for solving relaxations of discrete optimization problems. Currently included are solvers for discrete graphical models, graph matching and multicut. GitHub repository
Region Covariance Descriptors implemented in pure Matlab as described in "Region Covariance: A Fast Descriptor for Detection and Classification" by Oncel Tuzel, Fatih Porikli, and Peter Meer, ECCV 2006: Region covariance descriptor.