A Comparative Study of Energy Minimization Methods for Markov Random Fields with Smoothness-Based Priors

By Richard Szeliski, Ramin Zabih, Daniel Scharstein, Olga Veksler, Vladimir Kolmogorov, Aseem Agarwala, Marshall F. Tappen, Carsten Rother. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008

1,072 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/00c9b3dbd0a6a85bd201c7ee0823a388907a273d

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

DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMatch (2019, methodology): “cost aggregation, semi-global matching [14], and Markov random fields [30, 34], they are able to achieve good performance on relatively simple scenarios.”
Diverse M-Best Solutions in Markov Random Fields (2012, background): “Indeed, recent empirical studies [20, 32] have repeatedly found that MAP solutions of existing models are of much poorer quality than the ground-truth on vision problems like segmentation, stereo, optical flow, denoising, etc.”
Superpixels and Supervoxels in an Energy Optimization Framework (2010, methodology): “We optimize with graph cuts [18], which is known to perform well [19].”
OpenGM: A C++ Library for Discrete Graphical Models (2012, background): “MRF-lib (Szeliski et al., 2008) is restricted to the min-sum semi-ring and second-order grid graphs.”
PatchMatch: a randomized correspondence algorithm for structural image editing (2009, methodology): “The nature of our algorithm bears some superficial similarity to LBP and Graph Cuts algorithms often used to solve Markov Random Fields on an image grid [Szeliski et al. 2008].”
Mathematics for Machine Learning (2020, background): “, image segmentation, semantic labeling, image denoising, image restoration (Kittler and Föglein, 1984; Sucar and Gillies, 1994; Shotton et al., 2006; Szeliski et al., 2008)) to coding theory (McEliece et al.”
SIFT Flow: Dense Correspondence across Scenes and Its Applications (2011, methodology): “The fact that the underlying correspondence between scenes is complicated and unclear, and detecting contours from scenes can be unreliable, leads us to seek for dense, pixel-wise correspondence for scene alignment.”
Contextually guided semantic labeling and search for three-dimensional point clouds (2013, background): “Most of the works on 2D images have only used associative edge potentials (e.g., Szeliski et al., 2008), where the idea is that visual similarity is usually an indicator of whether two segments (or pixels) belong to same class.”
Beyond pixels: exploring new representations and applications for motion analysis (2009, methodology): “This energy function can be directly optimized by running sequential Belief Propagation (BP-S) [114] on a dual plane setup [103].”
On the Partition Function and Random Maximum A-Posteriori Perturbations (2012, background): “, 2009), stereo vision (Szeliski et al., 2007), parsing (Koo et al.”
Optical flow modeling and computation: A survey (2015, methodology): “…solution can be obtained much faster with the Iterated Conditional Modes [25] or its High Conﬁdence First variant [61], operating by iterative local minimizations of the energy, but this local optimum often yields poor results compared to modern methods [237], especially for non-convex functions.”
MRF Energy Minimization and Beyond via Dual Decomposition (2011, neutral): “…each node p 2 V such that the following objective function, known as the MRF energy, is minimized: X p2V p ðl p Þ þ X pq2E pq ðl p ; l q Þ: ð1Þ Currently, two classes of methods are the most prominent ones in MRF optimization [46], [47]: those based on graph-cuts and those based on message-passing.”
Multi-View Stereo: A Tutorial (2015, methodology): “MRF is a popular formulation for many other Computer Vision problems as well, and more details on MRF can be found in the following articles [114, 179].”
A Fully-Connected Layered Model of Foreground and Background Flow (2013, background): “Graph cuts (GC) [4] and belief propagation (BP) [17] are popular optimizers for Ising/Potts models and can find better local optima than local search methods [28].”