Multi-Dimensional Edge Features Graph Neural Network On Few-Shot Image Classification

Few-shot image classification with graph neural network (GNN) is a hot topic in recent years. Most GNN-based approaches have achieved promising performance. These methods utilize node features or one-dimensional edge feature for classification ignoring rich edge featues between nodes. In this letter, we propose a novel graph neural network exploiting multi-dimensional edge features (MDE-GNN) based on edge-labeling graph neural network (EGNN) and transductive neural network for few-shot learning. Unlike previous GNN-based approaches, we utilize multi-dimensional edge features information to construct edge matrices in graph. After layers of node and edge feautres updating, we generate a similarity score matrix by the mulit-dimensional edge features through a well-designed edge aggregation module. The parameters in our network are iteratively learnt by episode training with an edge similarity loss. We apply our model to supervised few-shot image classification tasks. Compared with previous GNNs and other few-shot learning approaches, we achieve state-of-the-art performance with two benchmark datasets.