A Task-Aware Dual Similarity Network for Fine-Grained Few-Shot Learning

AbstractThe goal of fine-grained few-shot learning is to recognize sub-categories under the same super-category by few labeled samples. Most recent approaches adopt a single similarity measure, that is, global or local measure alone. However, for images with high intra-class variance and low inter-class variance, exploring invariant features discriminative details quite essential. In this paper, we propose Task-aware Dual Similarity Network (TDSNet), which applies patches achieve better performance. Specifically, feature enhancement module adopted activate strong discriminability. Besides, task-aware attention exploits important among entire task. Finally, both class prototypes obtained are employed prediction. Extensive experiments on three datasets demonstrate proposed TDSNet achieves competitive performance comparing other state-of-the-art algorithms.KeywordsFine-grained image classificationFew-shot learningFeature