Skip to Main content Skip to Navigation
New interface
Conference papers

Towards Good Practice in Large-Scale Learning for Image Classification

Florent Perronnin 1 Zeynep Akata 1, 2 Zaid Harchaoui 2 Cordelia Schmid 2 
2 LEAR - Learning and recognition in vision
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann, Grenoble INP - Institut polytechnique de Grenoble - Grenoble Institute of Technology
Abstract : We propose a benchmark of several objective functions for large-scale image classification: we compare the one- vs-rest, multiclass, ranking and weighted average ranking SVMs. Using stochastic gradient descent optimization, we can scale the learning to millions of images and thousands of classes. Our experimental evaluation shows that ranking based algorithms do not outperform a one-vs-rest strategy and that the gap between the different algorithms reduces in case of high-dimensional data. We also show that for one-vs-rest, learning through cross-validation the optimal degree of imbalance between the positive and the negative samples can have a significant impact. Furthermore, early stopping can be used as an effective regularization strategy when training with stochastic gradient algorithms. Follow- ing these "good practices", we were able to improve the state-of-the-art on a large subset of 10K classes and 9M of images of ImageNet from 16.7% accuracy to 19.1%.
Document type :
Conference papers
Complete list of metadata

Cited literature [38 references]  Display  Hide  Download
Contributor : THOTH Team Connect in order to contact the contributor
Submitted on : Friday, April 20, 2012 - 6:39:35 PM
Last modification on : Saturday, November 19, 2022 - 3:58:49 AM
Long-term archiving on: : Saturday, July 21, 2012 - 2:35:37 AM


Files produced by the author(s)



Florent Perronnin, Zeynep Akata, Zaid Harchaoui, Cordelia Schmid. Towards Good Practice in Large-Scale Learning for Image Classification. CVPR 2012 - IEEE Computer Vision and Pattern Recognition, Jun 2012, Providence (RI), United States. pp.3482-3489, ⟨10.1109/CVPR.2012.6248090⟩. ⟨hal-00690014⟩



Record views


Files downloads