Closed-form approximate CRF training for scalable image segmentation

A. Kolesnikov, M. Guillaumin, V. Ferrari, C. Lampert, in:, D. Fleet, T. Pajdla, B. Schiele, T. Tuytelaars (Eds.), Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Springer, 2014, pp. 550–565.


Conference Paper | Published | English
Author
; ; ;
Editor
; ; ;
Department
Series Title
LNCS
Abstract
We present LS-CRF, a new method for training cyclic Conditional Random Fields (CRFs) from large datasets that is inspired by classical closed-form expressions for the maximum likelihood parameters of a generative graphical model with tree topology. Training a CRF with LS-CRF requires only solving a set of independent regression problems, each of which can be solved efficiently in closed form or by an iterative solver. This makes LS-CRF orders of magnitude faster than classical CRF training based on probabilistic inference, and at the same time more flexible and easier to implement than other approximate techniques, such as pseudolikelihood or piecewise training. We apply LS-CRF to the task of semantic image segmentation, showing that it achieves on par accuracy to other training techniques at higher speed, thereby allowing efficient CRF training from very large training sets. For example, training a linearly parameterized pairwise CRF on 150,000 images requires less than one hour on a modern workstation.
Publishing Year
Date Published
2014-09-01
Proceedings Title
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume
8691
Issue
PART 3
Page
550 - 565
Conference
ECCV: European Conference on Computer Vision
Conference Location
Zurich, Switzerland
Conference Date
2014-09-06 – 2014-09-12
IST-REx-ID

Cite this

Kolesnikov A, Guillaumin M, Ferrari V, Lampert C. Closed-form approximate CRF training for scalable image segmentation. In: Fleet D, Pajdla T, Schiele B, Tuytelaars T, eds. Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol 8691. Springer; 2014:550-565. doi:10.1007/978-3-319-10578-9_36
Kolesnikov, A., Guillaumin, M., Ferrari, V., & Lampert, C. (2014). Closed-form approximate CRF training for scalable image segmentation. In D. Fleet, T. Pajdla, B. Schiele, & T. Tuytelaars (Eds.), Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 8691, pp. 550–565). Zurich, Switzerland: Springer. https://doi.org/10.1007/978-3-319-10578-9_36
Kolesnikov, Alexander, Matthieu Guillaumin, Vittorio Ferrari, and Christoph Lampert. “Closed-Form Approximate CRF Training for Scalable Image Segmentation.” In Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), edited by David Fleet, Tomas Pajdla, Bernt Schiele, and Tinne Tuytelaars, 8691:550–65. Springer, 2014. https://doi.org/10.1007/978-3-319-10578-9_36.
A. Kolesnikov, M. Guillaumin, V. Ferrari, and C. Lampert, “Closed-form approximate CRF training for scalable image segmentation,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Zurich, Switzerland, 2014, vol. 8691, no. PART 3, pp. 550–565.
Kolesnikov A, Guillaumin M, Ferrari V, Lampert C. 2014. Closed-form approximate CRF training for scalable image segmentation. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). ECCV: European Conference on Computer Vision, LNCS, vol. 8691. 550–565.
Kolesnikov, Alexander, et al. “Closed-Form Approximate CRF Training for Scalable Image Segmentation.” Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), edited by David Fleet et al., vol. 8691, no. PART 3, Springer, 2014, pp. 550–65, doi:10.1007/978-3-319-10578-9_36.

Link(s) to Main File(s)
Access Level
OA Open Access

Export

Marked Publications

Open Data IST Research Explorer

Search this title in

Google Scholar