Mesh Model Codec Based on Fusion Graph Convolution and Parallel Pooling

Three-dimensional shape generation is an important research direction of computer graphics. In this paper, a graph convolutional neural network is introduced to obtain local features for a three-dimensional mesh model data set with the same connectivity relationship. A parallel pooling function is proposed to jointly extract richer global features, improve network generation performance, and ensure the robustness of the model during network training. Experiments are carried out on datasets such as Swing. The experimental results show that the mesh model codec network based on fusion graph convolution and parallel pooling proposed in this paper can effectively obtain global and local features in grid data and achieve a higher-quality mesh generation effect.