A Broad Study on the Transferability of Visual Representations with Contrastive Learning

Tremendous progress has been made in visual representation learning, notably with the recent success of self-supervised contrastive learning methods. Supervised contrastive learning has also been shown to outperform its cross-entropy counterparts by leveraging labels for choosing where to contrast. However, there has been little work to explore the transfer capability of contrastive learning to a different domain. In this paper, we conduct a comprehensive study on the transferability of learned representations of different contrastive approaches for linear evaluation, full-network transfer, and few-shot recognition on 12 downstream datasets from different domains, and object detection tasks on MSCOCO and VOC0712. The results show that the contrastive approaches learn representations that are easily transferable to a different downstream task. We further observe that the joint objective of self-supervised contrastive loss with cross-entropy/supervised-contrastive loss leads to better transferability of these models over their supervised counterparts. Our analysis reveals that the representations learned from the contrastive approaches contain more low/mid-level semantics than cross-entropy models, which enables them to quickly adapt to a new task. Our codes and models will be publicly available to facilitate future research on transferability of visual representations.¹

1. Introduction

Self-supervised learning is an important research area whose goal is to learn superior data representations without any labelled supervision. Recently, self-supervised contrastive learning has shown promising results in image classification tasks [21], [7], [4]. In the contrastive learning paradigm, a model is trained to recognize different augmentations of the same image (commonly referred as positives) while discriminating them from other random images (referred as negatives) in the dataset. The promising performance of self-supervised contrastive learning led to the idea of leveraging label information in the contrastive learning paradigm. To this end, Khosla et al. [30] proposed a supervised contrastive learning framework that achieves better ImageNet accuracy than the standard cross-entropy model.