Efficient Supervised Image Clustering Based on Density Division and Graph Neural Networks

In recent research, supervised image clustering based on Graph Neural Networks (GNN) connectivity prediction has demonstrated considerable improvements over traditional algorithms. However, existing algorithms are usually time-consuming and limit their applications. order to infer the between instances, they created a subgraph for each instance. Due creation process of large number subgraphs as input GNN, computation overheads enormous. To address high overhead problem in GNN prediction, we present time-efficient effective GNN-based framework density division namely DDC-GNN. DDC-GNN divides all instances into high-density parts low-density parts, only performs resulting significant reduction redundant calculations. We test two typical models module framework, which graph convolutional networks (GCN)-based model auto-encoder (GAE)-based model. Meanwhile, adaptive generated ensure sufficient contextual information extraction instead fixed-size subgraphs. According experiments different datasets, achieves higher accuracy is almost five times quicker than those without strategy.