A Comparative Study of Energy Minimization Methods for Markov Random Fields

By Richard Szeliski, Ramin Zabih, Daniel Scharstein, Olga Veksler, Vladimir Kolmogorov, Aseem Agarwala, Marshall F. Tappen, Carsten Rother. European Conference on Computer Vision, 2006

478 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/5fe100d7e7501410d6ed16bd3e9cf17872fd783e

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

Citations in the corpus

Optimizing Binary MRFs via Extended Roof Duality (2007, methodology): “Finally, we will make the code and energies publicly available, as a step towards a benchmarking system for optimizing challenging, non-submodular MRFs in computer vision, similar to [22].”
Natural Image Denoising with Convolutional Networks (2008, background): “As a result, a great deal of research has been devoted to approximate MRF learning and inference techniques that meliorate computational difficulties, generally at the cost of either representational power or theoretical guarantees [12] [13].”
Modeling the World from Internet Photo Collections (2008, background): “There is also a need for evaluation studies, in the spirit of (Scharstein and Szeliski 2002; Seitz et al. 2006; Szeliski et al. 2006), that benchmark the best-of-breed SfM algorithms against ground truth datasets and encour-”
Evaluation of Cost Functions for Stereo Matching (2007, methodology): “Our implementation is based on the MRF library provided by [19].”
Asymmetric Cuts: Joint Image Labeling and Partitioning (2014, background): “Many common pixel or superpixel labeling problems [12,33,35,36,6,17] are then formulated as an energy minimization over the sums of unary and pairwise terms:”
Graphical Models, Exponential Families, and Variational Inference (2008, methodology): “Various forms of these reweighted maxproduct algorithms have been applied in problems such as segmentation and disparity problems in computer vision [165, 134, 136, 246, 222, 260], error-control coding [73], side-chain prediction [261, 246], sensor fusion [43, 46].”
Get Out of my Picture! Internet-based Inpainting (2009, methodology): “To optimise the MRF described above, we use tree-reweighted belief propagation [18, 27, 28], using the software made available online [2] by the authors of [27].”
Subgradient Methods for Structured Prediction (2007, methodology): “This computation time is primarily dominated by executing of the alpha-beta expansion algorithm (Szeliski et al., 2006).”
Contextual classification with functional Max-Margin Markov Networks (2009, background): “Since max-margin learning does not require computation of marginals, we can learn by using efficient graph-cut inference, which are typically faster while as accurate as belief propagation and variants [23].”
SKYLINE2GPS: Localization in urban canyons using omni-skylines (2010, methodology): “In our problem we use graph cuts, which is both fast and highly successful in various vision problems like stereo, segmentation, and image restoration [29].”
Automatic Image Colorization Via Multimodal Predictions (2008, methodology): “We use the graph cut package(7) provided by [13].”
GeoS: Geodesic Image Segmentation (2008, methodology): “VGA sized); (b) unaries computed from the user scribbles provided in [8]; (c) GeoS segmentations.”
Multi-Class Segmentation with Relative Location Prior (2008, background): “[27] compared different energy minimization (max-assignment inference) methods and found that such methods are superior for the binary image segmentation task.”
Graph Cut Based Inference with Co-occurrence Statistics (2010, background): “As the memory requirements of any conventional inference algorithm [30] is typically O(|V|) for vision problems, the memory requirements of a formulation incorporating co-occurrence potentials should also be O(|V|).”
A Linear Programming Approach to Max-Sum Problem: A Review (2007, background): “The task is different from nonsupermodular problems of the Potts type and arising from stereo reconstruction, experimentally examined in [19], [29], [30], [32], [72], [73], in the fact that a lot of edge qualities are 1.”
Learning Conditional Random Fields for Stereo (2007, methodology): “However, here we use graph cuts for approximate MPE inference since it is faster, and also because the results of [13] suggest that graph cuts finds lower energy solutions than BP.”
3-D Video Representation Using Depth Maps (2011, methodology): “Usually, depth estimation algorithms attempt to match corresponding signal components in two or more original cameras, using a matching function [ 44 ] with different area support and size [3].”