2021/06/29 by Wenming Tang Guoping Qiu, Qiu, Wenming Tang Guoping · 1 citation
Computer Science · #Advanced Neural Network Applications #Video Surveillance and Tracking Methods #Human Pose and Action Recognition
paper · pdf · doi:10.48550/arxiv.2106.15778
This paper presents new designs of graph convolutional neural networks (GCNs)\non 3D meshes for 3D object segmentation and classification. We use the faces of\nthe mesh as basic processing units and represent a 3D mesh as a graph where\neach node corresponds to a face. To enhance the descriptive power of the graph,\nwe introduce a 1-ring face neighbourhood structure to derive novel\nmulti-dimensional spatial and structure features to represent the graph nodes.\nBased on this new graph representation, we then design a densely connected\ngraph convolutional block which aggregates local and regional features as the\nkey construction component to build effective and efficient practical GCN\nmodels for 3D object classification and segmentation. We will present\nexperimental results to show that our new technique outperforms state of the\nart where our models are shown to have the smallest number of parameters and\nconsietently achieve the highest accuracies across a number of benchmark\ndatasets. We will also present ablation studies to demonstrate the soundness of\nour design principles and the effectiveness of our practical models.\n