Large Scale Visual Classification with Parallel, Imbalanced Bagging of Incremental LIBLINEAR SVM

An ImageNet dataset with more than 14M images and 21K classes makes the problem of visual classification more difficult to deal with. One of the most difficult tasks is to train a fast and accurate classifier. In this paper, the authors address this challenge by extending the state-of-the-art large...

Full description

Saved in:
Bibliographic Details
Published inProceedings of the International Conference on Data Mining (DMIN) p. 1
Main Authors Doan, Thanh-Nghi, Do, Thanh-Nghi, Poulet, François
Format Conference Proceeding
LanguageEnglish
Published Athens The Steering Committee of The World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp) 01.01.2013
Online AccessGet full text

Cover

More Information
Summary:An ImageNet dataset with more than 14M images and 21K classes makes the problem of visual classification more difficult to deal with. One of the most difficult tasks is to train a fast and accurate classifier. In this paper, the authors address this challenge by extending the state-of-the-art large scale linear classifier LIBLINEAR-CDBLOCK proposed by Hsiang-Fu Yu in three ways: 1. improve LIBLINEARCDBLOCK for large number of classes with one-versus-all approach, 2. a balanced bagging algorithm for training binary classifiers, and 3. parallelize the training process of classifiers with several multi-core computers. The approach is evaluated on the 100 largest classes of an ImageNet and ILSVRC 2010. The evaluation shows that the approach is 732 times faster than the original implementation and 1193 times faster than LIBLINEAR without (or very few) compromising classification accuracy.