{"_id":"1859","status":"public","year":"2015","citation":{"ieee":"N. Shah, V. Kolmogorov, and C. Lampert, “A multi-plane block-coordinate Frank-Wolfe algorithm for training structural SVMs with a costly max-oracle,” presented at the CVPR: Computer Vision and Pattern Recognition, Boston, MA, USA, 2015, pp. 2737–2745.","chicago":"Shah, Neel, Vladimir Kolmogorov, and Christoph Lampert. “A Multi-Plane Block-Coordinate Frank-Wolfe Algorithm for Training Structural SVMs with a Costly Max-Oracle,” 2737–45. IEEE, 2015. <a href=\"https://doi.org/10.1109/CVPR.2015.7298890\">https://doi.org/10.1109/CVPR.2015.7298890</a>.","ista":"Shah N, Kolmogorov V, Lampert C. 2015. A multi-plane block-coordinate Frank-Wolfe algorithm for training structural SVMs with a costly max-oracle. CVPR: Computer Vision and Pattern Recognition, 2737–2745.","apa":"Shah, N., Kolmogorov, V., &#38; Lampert, C. (2015). A multi-plane block-coordinate Frank-Wolfe algorithm for training structural SVMs with a costly max-oracle (pp. 2737–2745). Presented at the CVPR: Computer Vision and Pattern Recognition, Boston, MA, USA: IEEE. <a href=\"https://doi.org/10.1109/CVPR.2015.7298890\">https://doi.org/10.1109/CVPR.2015.7298890</a>","ama":"Shah N, Kolmogorov V, Lampert C. A multi-plane block-coordinate Frank-Wolfe algorithm for training structural SVMs with a costly max-oracle. In: IEEE; 2015:2737-2745. doi:<a href=\"https://doi.org/10.1109/CVPR.2015.7298890\">10.1109/CVPR.2015.7298890</a>","short":"N. Shah, V. Kolmogorov, C. Lampert, in:, IEEE, 2015, pp. 2737–2745.","mla":"Shah, Neel, et al. <i>A Multi-Plane Block-Coordinate Frank-Wolfe Algorithm for Training Structural SVMs with a Costly Max-Oracle</i>. IEEE, 2015, pp. 2737–45, doi:<a href=\"https://doi.org/10.1109/CVPR.2015.7298890\">10.1109/CVPR.2015.7298890</a>."},"scopus_import":1,"day":"01","project":[{"grant_number":"308036","_id":"2532554C-B435-11E9-9278-68D0E5697425","name":"Lifelong Learning of Visual Scene Understanding","call_identifier":"FP7"},{"grant_number":"616160","_id":"25FBA906-B435-11E9-9278-68D0E5697425","name":"Discrete Optimization in Computer Vision: Theory and Practice","call_identifier":"FP7"}],"type":"conference","ec_funded":1,"publist_id":"5240","main_file_link":[{"open_access":"1","url":"http://arxiv.org/abs/1408.6804"}],"publisher":"IEEE","quality_controlled":"1","oa_version":"Preprint","oa":1,"author":[{"last_name":"Shah","full_name":"Shah, Neel","first_name":"Neel","id":"31ABAF80-F248-11E8-B48F-1D18A9856A87"},{"id":"3D50B0BA-F248-11E8-B48F-1D18A9856A87","last_name":"Kolmogorov","first_name":"Vladimir","full_name":"Kolmogorov, Vladimir"},{"id":"40C20FD2-F248-11E8-B48F-1D18A9856A87","orcid":"0000-0001-8622-7887","full_name":"Lampert, Christoph","first_name":"Christoph","last_name":"Lampert"}],"publication_status":"published","month":"06","language":[{"iso":"eng"}],"doi":"10.1109/CVPR.2015.7298890","title":"A multi-plane block-coordinate Frank-Wolfe algorithm for training structural SVMs with a costly max-oracle","department":[{"_id":"VlKo"},{"_id":"ChLa"}],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","abstract":[{"text":"Structural support vector machines (SSVMs) are amongst the best performing models for structured computer vision tasks, such as semantic image segmentation or human pose estimation. Training SSVMs, however, is computationally costly, because it requires repeated calls to a structured prediction subroutine (called \\emph{max-oracle}), which has to solve an optimization problem itself, e.g. a graph cut.\r\nIn this work, we introduce a new algorithm for SSVM training that is more efficient than earlier techniques when the max-oracle is computationally expensive, as it is frequently the case in computer vision tasks. The main idea is to (i) combine the recent stochastic Block-Coordinate Frank-Wolfe algorithm with efficient hyperplane caching, and (ii) use an automatic selection rule for deciding whether to call the exact max-oracle or to rely on an approximate one based on the cached hyperplanes.\r\nWe show experimentally that this strategy leads to faster convergence to the optimum with respect to the number of requires oracle calls, and that this translates into faster convergence with respect to the total runtime when the max-oracle is slow compared to the other steps of the algorithm. ","lang":"eng"}],"date_published":"2015-06-01T00:00:00Z","page":"2737 - 2745","date_created":"2018-12-11T11:54:24Z","date_updated":"2021-01-12T06:53:40Z","conference":{"end_date":"2015-06-12","location":"Boston, MA, USA","start_date":"2015-06-07","name":"CVPR: Computer Vision and Pattern Recognition"}}