An empirical study of pretrained representations for few-shot classification
This work addresses the problem of selecting optimal pretrained models for few-shot classification, which is important for researchers and practitioners in computer vision, but it is incremental as it focuses on empirical comparison rather than introducing new methods.
The paper systematically investigates which pretrained models provide the best representations for few-shot image classification, finding that supervised models outperform unsupervised ones, even on out-of-distribution datasets, and that adversarially robust models transfer better but with slightly lower accuracy.
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.