Breadcrumbs: Adversarial Class-Balanced Sampling for Long-tailed Recognition

TL;DR

This paper introduces Breadcrumb, an adversarial class-balanced sampling method combined with EMANATE feature augmentation, to improve long-tailed recognition performance without overfitting to few-shot classes.

Contribution

It proposes Breadcrumb sampling and EMANATE augmentation, providing a novel adversarial approach to address overfitting in long-tailed recognition tasks.

Findings

Outperforms existing long-tailed recognition methods on multiple datasets.

Effectively mitigates overfitting to few-shot classes.

Enhances classifier robustness in imbalanced data scenarios.

Abstract

The problem of long-tailed recognition, where the number of examples per class is highly unbalanced, is considered. While training with class-balanced sampling has been shown effective for this problem, it is known to over-fit to few-shot classes. It is hypothesized that this is due to the repeated sampling of examples and can be addressed by feature space augmentation. A new feature augmentation strategy, EMANATE, based on back-tracking of features across epochs during training, is proposed. It is shown that, unlike class-balanced sampling, this is an adversarial augmentation strategy. A new sampling procedure, Breadcrumb, is then introduced to implement adversarial class-balanced sampling without extra computation. Experiments on three popular long-tailed recognition datasets show that Breadcrumb training produces classifiers that outperform existing solutions to the problem.