EIFA-KD: Explicit and implicit feature augmentation with knowledge distillation for long-tailed visual data classification

Xiyan Deng

doi:10.1016/j.patcog.2025.112129

EIFA-KD: Explicit and implicit feature augmentation with knowledge distillation for long-tailed visual data classification

2025-07-11

0

PRE

AI

DOI：10.1016/j.patcog.2025.112129

Original

Original request for help

Save

Abstract

En 中文

• Propose an explicit feature augmentation method for generating new samples of tail classes. • Propose an implicit feature enhancement method to calibrate the distribution of tail class. • A multi-branch network based on knowledge distillation is designed to generate high-quality features by aggregating the knowledge of multi-sub-branch networks.

Keywords:

feature augmentation

feature enhancement

multi-branch network

knowledge distillation

tail class learning

EIFA-KD: Explicit and implicit feature augmentation with knowledge distillation for long-tailed visual data classification

Abstract

Journal

Organization