Adjusting Logit in Gaussian Form for Long-Tailed Visual Recognition

Abstract

It is not uncommon that real-world data are distributed with a long tail. For such data, the learning of deep neural networks becomes challenging because it is hard to classify tail classes correctly. In the literature, several existing methods have addressed this problem by reducing classifier bias, provided that the features obtained with long-tailed data are representative enough. However, we find that training directly on long-tailed data leads to uneven embedding space. That is, the embedding space of head classes severely compresses that of tail classes, which is not conducive to subsequent classifier learning. This paper therefore studies the problem of long-tailed visual recognition from the perspective of feature level. We introduce feature augmentation to balance the embedding distribution. The features of different classes are perturbed with varying amplitudes in Gaussian form. Based on these perturbed features, two novel logit adjustment methods are proposed to improve model performance at a modest computational overhead. Subsequently, the distorted embedding spaces of all classes can be calibrated. In such balanced-distributed embedding spaces, the biased classifier can be eliminated by simply retraining the classifier with classbalanced sampling data. Extensive experiments conducted on benchmark datasets demonstrate the superior performance of the proposed method over the state-of-the-art ones. Source code is available at https://github.com/Keke921/GCLLoss. Impact Statement-Long-tailed visual recognition, a burgeoning field within computer vision, holds profound significance in academic discourse. It fosters advancements in real-world applications by addressing challenges posed by imbalanced datasets, thereby facilitating improved model generalization. In this paper, we propose a simple yet effective logit adjustment method, applicable across different models. Our work provides comprehensive discussions of the proposed method for longtail learning, considering aspects of optimization and geometric interpretation. These discussions contribute to a deeper understanding of long-tail learning and a novel approach for enhancing generalization on the test set. In scholarly pursuits, long-tailed visual recognition underscores the necessity for nuanced and