Hierarchical Dynamic Graph Clustering Network
Jie Chen,Licheng Jiao,Xu Liu,Lingling Li,Fang Liu,Puhua Chen,Shuyuan Yang,Biao Hou
DOI: https://doi.org/10.1109/tkde.2023.3333529
IF: 9.235
2024-01-01
IEEE Transactions on Knowledge and Data Engineering
Abstract:Connections between visual components are ubiquitous. Graphs, as a highly flexible data structure, not only allow imposing relational induction bias on data, but can provide a completely distinct learning perspective for regular image data. In this article, we propose a hierarchical dynamic graph clustering network (HDGCN) for visual feature learning. We construct hierarchical graph representations in graph domain in an adaptive, data-adaptive and task-adaptive manner. First, the initial graph is constructed in high-dimensional feature domain of images. To mine the hierarchical geometric features in latent graph space, adaptive clustering network (ClusterNet) is performed to learn discriminative clusters and generates cluster-based coarse graph. Then, graph convolutional networks (GCNs) are used to diffuse, transform and aggregate information among clusters. So, the intra-class and inter-class information is fully explored to increase the discriminativity of graph representations. Next, coarsened graph representations are mapped to grid based on its affinity with linear projection features. To further improve the task adaptation of clusters and hierarchical graph representations, ClusterNet and GCNs are fused in the same framework for end-to-end training and clusters is updated dynamically. We have conducted extensive experiments on classification and segmentation tasks. The experimental results fully validate the robustness of the proposed algorithm.