A Fast Face Identification Method for Wireframe Model Fusing Graph Neural Network

In computer-aided design systems, when 3D reconstruction is performed from wireframe, the most critical step is to find the corresponding annular edges of 3D solid surfaces from the wireframe. Traditional face identification algorithms cannot effectively deal with nested loops and curve surface, as well as the large number of parameters and high computing resource consumption of existing deep learning methods. So, a lightweight face identification algorithms method called Edge-Sequence Graph Transformer (ES-GT) is proposed. Firstly, gated graph neural network is used to effectively capture the spatial dependence of edge nodes in the wireframe. Then, the virtual global node and the information fusion module are used to transmit messages to the disconnected nodes in the graph to achieve effective encoding, and finally the Transformer is used to decode and predict. Compared with the existing loop recognition algorithms, the precision and recall evaluation indexes of the proposed method are improved on the 2D and 3D wireframe datasets, and the algorithm requires fewer parameters and runs faster, the framework can effectively deal with the problem of loop identification in wireframe in real time.