Recent algorithms with state-of-the-art few-shot classification results start their procedure by computing data features output by a large pretrained model. In this paper we systematically investigate which models provide the best representations for a few-shot image classification task when pretrained on the Imagenet dataset. We test their representations when used as the starting point for different few-shot classification algorithms. We observe that models trained on a supervised classification task have higher performance than models trained in an unsupervised manner even when transferred to out-of-distribution datasets. Models trained with adversarial robustness transfer better, while having slightly lower accuracy than supervised models.
@article{arxiv.1910.01319,
title = {An empirical study of pretrained representations for few-shot classification},
author = {Tiago Ramalho and Thierry Sousbie and Stefano Peluchetti},
journal= {arXiv preprint arXiv:1910.01319},
year = {2019}
}