City Research Online

Abstract

Fine-grained few-shot image classification aims to classify images from previously unseen subcategories when only a limited number of labeled examples are available for training. Feature alignment methods have demonstrated strong performance by semantically aligning support and query features, enabling more accurate similarity comparisons between images and enhancing classification accuracy. However, the redundant background information makes feature alignment less effective at extracting sufficiently discriminative features to distinguish fine-grained images with subtle differences. To address this limitation, we propose a novel feature alignment and filtering network (FAFN) that can achieve visual semantic alignment and effective background filtering. Specifically, FAFN consists of two key components: a feature alignment module, which uses an attention mechanism to semantically align visual features between the query and support images, and a query-aware feature filtering module, which effectively filters out background and other non-discriminative features from the aligned query features. This design enables the model to better capture subtle and discriminative features, thereby enhancing the learned similarity metric. Experimental results on three fine-grained few-shot image classification datasets consistently demonstrate that FAFN outperforms other methods. Codes are available at: https://github.com/spraise/FAFN.

Publication Type:	Article
Additional Information:	© 2026 This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/
Publisher Keywords:	Few-shot learning, Fine-grained image classification, Feature alignment, Background filtering
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > TK Electrical engineering. Electronics Nuclear engineering
Departments:	Bayes Business School Bayes Business School > Faculty of Actuarial Science & Insurance
SWORD Depositor:	Symplectic Administrator

Export

Downloads