Efficient Structured Prediction with Latent Variables for General Graphical Models

By Alexander Gerhard Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun. 2012

72 citations recorded by the source corpus.

In this paper we propose a unified framework for structured prediction with latent variables which includes hidden conditional random fields and latent structured support vector machines as special cases. We describe a local entropy approximation for this general formulation using duality, and derive an efficient message passing algorithm that is guaranteed to converge. We demonstrate its effectiveness in the tasks of image segmentation as well as 3D indoor scene understanding from single images, showing that our approach is superior to latent structured support vector machines and hidden conditional random fields.

Subfields: Computer Vision and Pattern Recognition, Probabilistic & Bayesian Methods

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 Probabilistic & Bayesian Methods

Citations in the corpus

Aligning Books and Movies: Towards Story-Like Visual Explanations by Watching Movies and Reading Books (2015, background): “Since ground-truth is only available for a sparse set of shots, we regard the states of unobserved nodes as hidden variables and learn the CRF weights with [32].”
Learning to segment under various forms of weak supervision (2015, background): “An alternative are Markov random fields with latent variables denoting the super-pixels, and observed variables representing image tags [43, 30].”
What Are You Talking About? Text-to-Image Coreference (2014, methodology): “For learning the weights for each potential in our MRF, we use the primal-dual method of [11], specifically, we use the implementation of [26].”
Exploring Compositional High Order Pattern Potentials for Structured Output Learning (2013, background): “The difference between the left and right side is the KL divergence between q and p∗ where p∗(h|y) = exp ( 1 T (c+W y) h ) ∑ h exp ( 1 T (c+W y) h ) .”
Blending Learning and Inference in Conditional Random Fields (2016, methodology): “…(Yao et al., 2012), 3D scene understanding (Lin et al., 2013), shape reconstruction (Salzmann and Urtasun, 2012), indoor scene understanding (Schwing et al., 2012a; Schwing and Urtasun, 2012), depth estimation (Yamaguchi et al., 2012), flow estimation (Yamaguchi et al., 2013) and…”
A High Performance CRF Model for Clothes Parsing (2014, methodology): “To learn the weights, we use the primal-dual method of [30] (we use the implementation of [31]), shown to be more efficient than other structure prediction learning algorithms.”