TY - GEN

T1 - Efficient large-scale structured learning

AU - Branson, Steve

AU - Beijbom, Oscar

AU - Belongie, Serge

PY - 2013

Y1 - 2013

N2 - We introduce an algorithm, SVM-IS, for structured SVM learning that is computationally scalable to very large datasets and complex structural representations. We show that structured learning is at least as fast-and often much faster-than methods based on binary classification for problems such as deformable part models, object detection, and multiclass classification, while achieving accuracies that are at least as good. Our method allows problem-specific structural knowledge to be exploited for faster optimization by integrating with a user-defined importance sampling function. We demonstrate fast train times on two challenging large scale datasets for two very different problems: Image Net for multiclass classification and CUB-200-2011 for deformable part model training. Our method is shown to be 10-50 times faster than {SVM}{struct}for cost-sensitive multiclass classification while being about as fast as the fastest 1-vs-all methods for multiclass classification. For deformable part model training, it is shown to be 50-1000 times faster than methods based on SVM struct, mining hard negatives, and Pegasos-style stochastic gradient descent. Source code of our method is publicly available.

AB - We introduce an algorithm, SVM-IS, for structured SVM learning that is computationally scalable to very large datasets and complex structural representations. We show that structured learning is at least as fast-and often much faster-than methods based on binary classification for problems such as deformable part models, object detection, and multiclass classification, while achieving accuracies that are at least as good. Our method allows problem-specific structural knowledge to be exploited for faster optimization by integrating with a user-defined importance sampling function. We demonstrate fast train times on two challenging large scale datasets for two very different problems: Image Net for multiclass classification and CUB-200-2011 for deformable part model training. Our method is shown to be 10-50 times faster than {SVM}{struct}for cost-sensitive multiclass classification while being about as fast as the fastest 1-vs-all methods for multiclass classification. For deformable part model training, it is shown to be 50-1000 times faster than methods based on SVM struct, mining hard negatives, and Pegasos-style stochastic gradient descent. Source code of our method is publicly available.

KW - cost-sensitive SVM

KW - deformable part models

KW - object detection

KW - optimization

KW - structured learning

KW - sub-gradient

UR - http://www.scopus.com/inward/record.url?scp=84887346435&partnerID=8YFLogxK

U2 - 10.1109/CVPR.2013.236

DO - 10.1109/CVPR.2013.236

M3 - Conference article

AN - SCOPUS:84887346435

SP - 1806

EP - 1813

JO - I E E E Conference on Computer Vision and Pattern Recognition. Proceedings

JF - I E E E Conference on Computer Vision and Pattern Recognition. Proceedings

SN - 1063-6919

M1 - 6619080

T2 - 26th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2013

Y2 - 23 June 2013 through 28 June 2013

ER -