MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition

TLDR

Real‑world training data often follows a long‑tailed distribution, causing majority classes to dominate and degrading supervised learning performance. The study proposes to augment minority classes by learning transformed semantic directions through meta‑learning, improving upon the standard implicit semantic data augmentation (ISDA) which struggles with scarce minority data. During training, the augmentation strategy is dynamically optimized via a meta‑update step that minimizes loss on a small balanced validation set. Experiments on CIFAR‑LT, ImageNet‑LT, and iNaturalist 2017/2018 demonstrate that the proposed method outperforms baselines and effectively mitigates long‑tail imbalance.

Abstract

Real-world training data usually exhibits long-tailed distribution, where several majority classes have a significantly larger number of samples than the remaining minority classes. This imbalance degrades the performance of typical supervised learning algorithms designed for balanced training sets. In this paper, we address this issue by augmenting minority classes with a recently proposed implicit semantic data augmentation (ISDA) algorithm [37], which produces diversified augmented samples by translating deep features along many semantically meaningful directions. Importantly, given that ISDA estimates the class-conditional statistics to obtain semantic directions, we find it ineffective to do this on minority classes due to the insufficient training data. To this end, we propose a novel approach to learn transformed semantic directions with meta-learning automatically. In specific, the augmentation strategy during training is dynamically optimized, aiming to minimize the loss on a small balanced validation set, which is approximated via a meta update step. Extensive empirical results on CIFAR-LT-10/100, ImageNet-LT, and iNaturalist 2017/2018 validate the effectiveness of our method.