Balanced Contrastive Learning for Long-Tailed Visual Recognition

TLDR

Real‑world datasets are long‑tailed, causing cross‑entropy classifiers to under‑represent minority classes, and while supervised contrastive learning works on balanced data, it fails to form a regular simplex on long‑tailed data. This work proposes balanced contrastive learning (BCL) to correct the optimization of supervised contrastive learning and improve long‑tailed visual recognition. BCL introduces class‑averaging to balance negative‑class gradients and class‑complement to include all classes in every mini‑batch, and is applied within a two‑branch framework. The method achieves a regular‑simplex representation, aids cross‑entropy optimization, and yields competitive results on CIFAR‑10‑LT, CIFAR‑100‑LT, ImageNet‑LT, and iNaturalist2018.

Abstract

Real-world data typically follow a long-tailed distribution, where a few majority categories occupy most of the data while most minority categories contain a limited number of samples. Classification models minimizing crossentropy struggle to represent and classify the tail classes. Although the problem of learning unbiased classifiers has been well studied, methods for representing imbalanced data are under-explored. In this paper, we focus on representation learning for imbalanced data. Recently, supervised contrastive learning has shown promising performance on balanced data recently. However, through our theoretical analysis, we find that for long-tailed data, it fails to form a regular simplex which is an ideal geometric configuration for representation learning. To correct the optimization behavior of SCL and further improve the performance of long-tailed visual recognition, we propose a novel loss for balanced contrastive learning (BCL). Compared with SCL, we have two improvements in BCL: classaveraging, which balances the gradient contribution of negative classes; class-complement, which allows all classes to appear in every mini-batch. The proposed balanced contrastive learning (BCL) method satisfies the condition of forming a regular simplex and assists the optimization of cross-entropy. Equipped with BCL, the proposed two-branch framework can obtain a stronger feature representation and achieve competitive performance on long-tailed benchmark datasets such as CIFAR-10-LT, CIFAR-100-LT, ImageNet-LT, and iNaturalist2018.