DICS-Net: Dictionary-guided Implicit-Component-Supervision Network for Few-Shot Classification

The few-shot classification (FSC) task has recently been a hot research topic. It aims to address the classification problem with insufficient labeled data on a cross-category basis. Typically, researchers pre-train a feature extractor with base data, then use it to extract the features of novel data and recognize them. Notably, the novel set only has a few annotated samples and has non-overlapped categories from the base set, which leads to that the pre-trained feature extractor can not adapt to the novel data flawlessly. We dub this problem as Feature-Extractor-Maladaptive (FEM) problem. Starting from the root cause of this problem, this paper presents a new scheme, Dictionary-guided Implicit-Component-Supervision Network (DICS-Net), to improve the performance of FSC. We believe that although the categories of base and novel sets are different, the composition of the sample's components is similar. For example, both cats and dogs contain leg and head components. Actually, such entity components are intra-class stable. They have fine cross-category versatility and new category generalization. However, in many real-world scenarios, common information of different categories (such as cats and airplanes) is not easy to find, which hinders the possibility of modeling based on this assumption. Therefore, we first design a Dictionary-based Implicit-Component Generator (DICG) to mine common information of different sets; then construct an implicit-component-based auxiliary task to improve the adaptability of the feature extractor. We conduct experiments on three benchmark datasets (mini-ImageNet, tiered-ImageNet, and FC100). The improvements of $0.9\%$-$10.1\%$ compared with state-of-the-arts have evaluated the efficiency of our DICS-Net.