Tailoring Embedding Function to Heterogeneous Few-Shot Tasks by Global and Local Feature Adaptors

Few-Shot Learning (FSL) is essential for visual recognition. Many methods tackle this challenging problem via learning an embedding function from seen classes and transfer it to unseen with a few labeled instances. Researchers recently found beneficial incorporate task-specific feature adaptation into FSL models, which produces the most representative features each task. However, these ignore diversity of apply global transformation In paper, we propose Global Local Feature Adaptor (GLoFA), unifying framework that tailors instance representation specific tasks by local adaptors. We claim class-specific helps improve ability adaptor. masks tend capture sketchy patterns, while focus on detailed characteristics. A strategy measure relationship between instances adaptively based characteristics both endow GLoFA handle mix-grained tasks. outperforms other heterogeneous task distribution achieves competitive results benchmark datasets.