Global- and local-aware feature augmentation with semantic orthogonality for few-shot image classification

Boyao Shi, Wenbin Li, Jing Huo, Pengfei Zhu, Lei Wang, Yang Gao

Abstract: Highlights•A semantic orthogonal learning framework (SOLF) is proposed to obtain orthogonal and diverse feature channels, exploring how to learn and generate better features for FSL in a pure semantic-aware manner, achieving better performance.•A global- and local- aware feature augmentation (GLFA) method is proposed to augment features in terms of improving the diversity of the augmented samples and alleviating the overfitting problem.•Extensive experiments on four standard benchmarks show that the proposed method significantly outperforms the baseline methods on the 5-way and even large-way settings, indicating the effectiveness and practicality for real-word scenarios.