Local Matching Relation Network for Few-shot Affordance Classification

Affordance refers to the potential ”operational possibilities” that a target object exhibits in an unknown external environment. Affordance classification is to classify objects with the same function into a class and predict affordance labels, which is a very important skill in the field of intelligent service robots serving human users. Classification models based on deep convolutional neural networks rely on huge amounts of training data, and there is a lack of datasets with a large number of annotations available for affordance classification, which can easily lead to overfitting. Therefore, it is of practical interest for us to train a model that can be quickly adapted to new affordance classification tasks by a small number of samples. Based on this, this paper applies the few-shot learning technique and proposes a local matching relation network. The model is based on relation network, and we propose a multi-scale fusion feature extractor as an embedding module to extract more effective image feature vectors. A local matching module is designed to improve the problem of poor classification due to the large intra-class variation of affordance images. Meanwhile, we build a dataset for affordance classification and conduct some affordance experiments. The experimental results verify the effectiveness of the method in this paper.