Convergent Tree-Reweighted Message Passing for Energy Minimization

By Vladimir Kolmogorov. 2006

1,341 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Electrical and Electronic Engineering, Optimization & Training

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Electrical and Electronic Engineering

In Optimization & Training

Citations in the corpus

Barrier Frank-Wolfe for Marginal Inference (2015, methodology): “The models are close to submodular and the local relaxation is a good approximation to the marginal polytope.”
Efficient Likelihood Learning of a Generic CNN-CRF Model for Semantic Segmentation (2015, methodology): “For this, LP-relaxation-based solvers, such as TRW-S (Kolmogorov, 2006), are among the best and most popular solvers.”
Displets: Resolving stereo ambiguities using object knowledge (2015, methodology): “We leverage greedy max-product particle belief propagation (MP-PBP) [44,55] with sequential tree-reweighted message passing (TRW-S) [25] using 30 particles and 50 iterations to find an approximate solution.”
Image Reconstruction from Bag-of-Visual-Words (2014, background): “However, leaving restrictions Equation 1, 2, 3 out of consideration, it can be solved approximately by belief propagation [47], α-expansion [48], or sequential treereweighted message passing [49], which are not applicable under the restrictions Equation 1, 2, 3.”
Hinge-Loss Markov Random Fields and Probabilistic Soft Logic (2015, background): “Many DD algorithms use coordinate descent, such as TRW-S (Kolmogorov, 2006), MSD (Werner, 2007), MPLP (Globerson and Jaakkola, 2007), and ADLP (Meshi and Globerson, 2011).”
Higher-order clique reduction in binary graph cut (2009, background): “Such optimization schemes have become quite popular, largely owing to the success of optimization techniques such as graph cuts[5, 10, 14], belief propagation[7, 21], and tree-reweighted message passing[13].”
What, Where and How Many? Combining Object Detectors and CRFs (2010, background): “One of the main advantages of our framework is that the associated energy function can be solved efﬁciently using graph cut [29] based move making algorithms (which outperform message passing algorithms [30,31] for many vision problems).”
Category Modeling from Just a Single Labeling: Use Depth Information to Guide the Learning of 2D Models (2013, methodology): “Pure graph matching based on TRW-S [36] without learning is denoted by Matching+TRW-S.”
Accelerated training of conditional random fields with stochastic gradient methods (2006, background): “Although LBP can sometimes oscillate, convergent versions have been developed ( e.g. , Kolmogorov, 2004).”
Deformable Medical Image Registration: A Survey (2013, methodology): “Kwon et al. [442] used TRW-S to perform inference in a factor graph that models higher-order spatial smoothness constraints for image registration.”
Joint 3d Estimation of Vehicles and Scene Flow (2015, methodology): “In the inner loop, we employ sequential tree-reweighted message passing (TRW-S) (Kolmogorov, 2006) to infer an approximate solution given the current set of particles.”
Discriminative Models for Multi-Class Object Layout (2009, methodology): “We compare our greedy inference algorithm to two other approximate inference approaches: Loopy Belief Propagation (LBP) and Tree Re-Weighted Belief Propagation (TRW) (Wainwright et al. 2002; Kolmogorov 2006).”
A Tutorial on Dual Decomposition and Lagrangian Relaxation for Inference in Natural Language Processing (2012, methodology): “Following the work of Kolmogorov (2006), we use TRW-E to refer to the edge-based variant of TRW, and TRW-T to refer to the tree-based algorithm. Kolmogorov (2006) derives a further variant, TRW-S (the “S” refers to the sequential nature of the algorithm). All three algorithms—TRW-E, TRW-T, and TRW-S”