What makes instance discrimination good for transfer learning?

TL;DR

This paper investigates why instance discrimination pretraining excels in transfer learning, revealing that low- and mid-level features are crucial, and proposes ways to improve supervised pretraining by leveraging exemplar-based methods.

Contribution

It provides insights into the features that matter for transfer learning and suggests enhancements to supervised pretraining using exemplar-based approaches.

Findings

Low- and mid-level representations are key for transfer.

Intra-category invariance reduces transferability.

Supervised pretraining can be improved with exemplar-based methods.

Abstract

Contrastive visual pretraining based on the instance discrimination pretext task has made significant progress. Notably, recent work on unsupervised pretraining has shown to surpass the supervised counterpart for finetuning downstream applications such as object detection and segmentation. It comes as a surprise that image annotations would be better left unused for transfer learning. In this work, we investigate the following problems: What makes instance discrimination pretraining good for transfer learning? What knowledge is actually learned and transferred from these models? From this understanding of instance discrimination, how can we better exploit human annotation labels for pretraining? Our findings are threefold. First, what truly matters for the transfer is low-level and mid-level representations, not high-level representations. Second, the intra-category invariance enforced by the traditional supervised model weakens transferability by increasing task misalignment. Finally, supervised pretraining can be strengthened by following an exemplar-based approach without explicit constraints among the instances within the same category.