Improving Tail-Class Representation with Centroid Contrastive Learning

TL;DR

This paper introduces ICCL, a novel contrastive learning method that enhances tail-class representations in long-tailed datasets by interpolating images and using centroids, leading to improved classification accuracy.

Contribution

It proposes interpolative centroid contrastive learning (ICCL) that better captures tail-class features by interpolating images and leveraging class centroids, addressing representation learning challenges in long-tailed data.

Findings

Achieves 2.8% accuracy improvement on iNaturalist 2018.

Effective in improving tail-class representation learning.

Demonstrates significant gains on multiple benchmarks.

Abstract

In vision domain, large-scale natural datasets typically exhibit long-tailed distribution which has large class imbalance between head and tail classes. This distribution poses difficulty in learning good representations for tail classes. Recent developments have shown good long-tailed model can be learnt by decoupling the training into representation learning and classifier balancing. However, these works pay insufficient consideration on the long-tailed effect on representation learning. In this work, we propose interpolative centroid contrastive learning (ICCL) to improve long-tailed representation learning. ICCL interpolates two images from a class-agnostic sampler and a class-aware sampler, and trains the model such that the representation of the interpolative image can be used to retrieve the centroids for both source classes. We demonstrate the effectiveness of our approach on multiple long-tailed image classification benchmarks. Our result shows a significant accuracy gain of 2.8% on the iNaturalist 2018 dataset with a real-world long-tailed distribution.