Enhancing Contrastive Learning Inspired by the Philosophy of "The Blind Men and the Elephant"

TL;DR

This paper introduces JointCrop and JointBlur, novel data augmentation methods inspired by the story of the blind men and the elephant, which improve contrastive learning by leveraging the joint distribution of augmentation parameters.

Contribution

It is the first to explicitly incorporate the joint distribution of augmentation parameters into contrastive learning, enhancing various baseline methods without extra computational cost.

Findings

Significant performance improvements on multiple contrastive learning frameworks.

Effective generation of challenging positive pairs through joint augmentation parameters.

No additional computational overhead introduced.

Abstract

Contrastive learning is a prevalent technique in self-supervised vision representation learning, typically generating positive pairs by applying two data augmentations to the same image. Designing effective data augmentation strategies is crucial for the success of contrastive learning. Inspired by the story of the blind men and the elephant, we introduce JointCrop and JointBlur. These methods generate more challenging positive pairs by leveraging the joint distribution of the two augmentation parameters, thereby enabling contrastive learning to acquire more effective feature representations. To the best of our knowledge, this is the first effort to explicitly incorporate the joint distribution of two data augmentation parameters into contrastive learning. As a plug-and-play framework without additional computational overhead, JointCrop and JointBlur enhance the performance of SimCLR, BYOL, MoCo v1, MoCo v2, MoCo v3, SimSiam, and Dino baselines with notable improvements.