Computing Midcurve with Multi-Layer and Convolutional Neural Networks

Midcurve computation: the dimension reduction of 2D thin polygon shapes to 1D curves is important for several Computer-Aided Design (CAD) and Computer-Aided Engineering (CAE) applications, particularly in finite element analysis and robot path planning. We advance the state-of-the-art by building on previous work on MidcurveNN, through two new neural architectures: a multi-layer dense network and a CNN-based architecture with skip connections. Through comparative analysis, we demonstrate that while the dense network shows limited improvements, the CNN-based architecture with skip connections achieves a tenfold reduction in average loss compared to existing methods. The proposed approach improves the quality of the generated Midcurves.