Mix from Failure: Confusion-Pairing Mixup for Long-Tailed Recognition

TL;DR

This paper introduces Confusion-Pairing Mixup (CP-Mix), a novel data augmentation technique that improves long-tailed image recognition by focusing on confusion pairs and mitigating class imbalance effects.

Contribution

CP-Mix estimates confusion distributions and augments data from confusion pairs to enhance minority class recognition in long-tailed datasets.

Findings

CP-Mix outperforms existing methods in long-tailed recognition tasks.

CP-Mix effectively reduces classifier confusion and bias.

Extensive experiments validate the superiority of CP-Mix.

Abstract

Long-tailed image recognition is a computer vision problem considering a real-world class distribution rather than an artificial uniform. Existing methods typically detour the problem by i) adjusting a loss function, ii) decoupling classifier learning, or iii) proposing a new multi-head architecture called experts. In this paper, we tackle the problem from a different perspective to augment a training dataset to enhance the sample diversity of minority classes. Specifically, our method, namely Confusion-Pairing Mixup (CP-Mix), estimates the confusion distribution of the model and handles the data deficiency problem by augmenting samples from confusion pairs in real-time. In this way, CP-Mix trains the model to mitigate its weakness and distinguish a pair of classes it frequently misclassifies. In addition, CP-Mix utilizes a novel mixup formulation to handle the bias in decision boundaries that originated from the imbalanced dataset. Extensive experiments demonstrate that CP-Mix outperforms existing methods for long-tailed image recognition and successfully relieves the confusion of the classifier.