NeurIPS 2019poster694 citations

Fixing the train-test resolution discrepancy

Hugo Touvron, Andrea Vedaldi, Matthijs Douze, Herve Jegou

Abstract

Data-augmentation is key to the training of neural networks for image classification. This paper first shows that existing augmentations induce a significant discrepancy between the size of the objects seen by the classifier at train and test time: in fact, a lower train resolution improves the classification at test time!

BibTeX
@inproceedings{NEURIPS2019_d03a857a,
 author = {Touvron, Hugo and Vedaldi, Andrea and Douze, Matthijs and Jegou, Herve},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {H. Wallach and H. Larochelle and A. Beygelzimer and F. d\textquotesingle Alch\'{e}-Buc and E. Fox and R. Garnett},
 pages = {},
 publisher = {Curran Associates, Inc.},
 title = {Fixing the train-test resolution discrepancy},
 url = {https://proceedings.neurips.cc/paper_files/paper/2019/file/d03a857a23b5285736c4d55e0bb067c8-Paper.pdf},
 volume = {32},
 year = {2019}
}
Fixing the train-test resolution discrepancy · NeurIPS 2019