Self-supervised visual representation learning has seen huge progress recently, but no large scale evaluation has compared the many models now available. We evaluate the transfer performance of 13 top self-supervised models on 40 downstream tasks, including many-shot and few-shot recognition, object detection, and dense prediction. We compare their performance to a supervised baseline and show that on most tasks the best self-supervised models outperform supervision, confirming the recently observed trend in the literature. We find ImageNet Top-1 accuracy to be highly correlated with transfer to many-shot recognition, but increasingly less so for few-shot, object detection and dense prediction. No single self-supervised method dominates overall, suggesting that universal pre-training is still unsolved. Our analysis of features suggests that top self-supervised learners fail to preserve colour information as well as supervised alternatives, but tend to induce better classifier calibration, and less attentive overfitting than supervised learners.
@article{arxiv.2011.13377,
title = {How Well Do Self-Supervised Models Transfer?},
author = {Linus Ericsson and Henry Gouk and Timothy M. Hospedales},
journal= {arXiv preprint arXiv:2011.13377},
year = {2021}
}
Comments
CVPR 2021. Code available at https://github.com/linusericsson/ssl-transfer