Deep Graph Matching with Improved Graph-Context Networks based by Attention
Shuowei Wang, Sifan Ding, William Zhu · 2024
Recently, in the study of graph matching problems, deep learning has gained increasing attention and rapid development. Although GCN is widely used in the vertex feature extraction stage, GCN has limited learning ability of vertex features, which affects the similarity measurement of vertexs, resulting in poor accuracy of the final solution. To solve this problem, a depth map matching model based on space encoder improved GCAN is proposed. The model uses the GCAN module to learn vertex features in the vertex feature extraction stage, and introduces a space encoder to learn the space structure of vertexs. In addition, an advanced combinatorial optimization solver is applied to solve the graph matching optimization to improve the flexibility of the model. The experiment results show that the accuracy of the model is improved on PASCALVOC data and Willow Object.