Supervised Contrastive Learning

TL;DR

This paper extends contrastive learning to the supervised setting, leveraging label information to improve image classification accuracy, robustness, and stability, with state-of-the-art results on ImageNet using ResNet-200.

Contribution

It introduces a supervised contrastive loss that effectively utilizes label information, outperforming traditional methods and enhancing robustness and stability.

Findings

Achieved 81.4% top-1 accuracy on ImageNet with ResNet-200.

Outperformed cross-entropy loss on multiple datasets and ResNet variants.

Improved robustness to natural image corruptions and hyperparameter stability.

Abstract

Contrastive learning applied to self-supervised representation learning has seen a resurgence in recent years, leading to state of the art performance in the unsupervised training of deep image models. Modern batch contrastive approaches subsume or significantly outperform traditional contrastive losses such as triplet, max-margin and the N-pairs loss. In this work, we extend the self-supervised batch contrastive approach to the fully-supervised setting, allowing us to effectively leverage label information. Clusters of points belonging to the same class are pulled together in embedding space, while simultaneously pushing apart clusters of samples from different classes. We analyze two possible versions of the supervised contrastive (SupCon) loss, identifying the best-performing formulation of the loss. On ResNet-200, we achieve top-1 accuracy of 81.4% on the ImageNet dataset, which is 0.8% above the best number reported for this architecture. We show consistent outperformance over cross-entropy on other datasets and two ResNet variants. The loss shows benefits for robustness to natural corruptions and is more stable to hyperparameter settings such as optimizers and data augmentations. Our loss function is simple to implement, and reference TensorFlow code is released at https://t.ly/supcon.