Leveraging Angular Information Between Feature and Classifier for Long-tailed Learning: A Prediction Reformulation Approach

TL;DR

This paper introduces an angle-based prediction reformulation for long-tailed learning, improving recognition performance without classifier re-balancing by leveraging angular relationships between features and classifiers.

Contribution

It proposes a novel angle-based prediction method that enhances long-tailed learning performance without re-balancing classifier weights, and explores modules to improve two-stage learning frameworks.

Findings

Achieves state-of-the-art results on CIFAR10/100-LT and ImageNet-LT

Outperforms existing methods without classifier re-balancing

Demonstrates the effectiveness of angular prediction in long-tailed scenarios

Abstract

Deep neural networks still struggle on long-tailed image datasets, and one of the reasons is that the imbalance of training data across categories leads to the imbalance of trained model parameters. Motivated by the empirical findings that trained classifiers yield larger weight norms in head classes, we propose to reformulate the recognition probabilities through included angles without re-balancing the classifier weights. Specifically, we calculate the angles between the data feature and the class-wise classifier weights to obtain angle-based prediction results. Inspired by the performance improvement of the predictive form reformulation and the outstanding performance of the widely used two-stage learning framework, we explore the different properties of this angular prediction and propose novel modules to improve the performance of different components in the framework. Our method is able to obtain the best performance among peer methods without pretraining on CIFAR10/100-LT and ImageNet-LT. Source code will be made publicly available.