2D B-spline Curve Reconstruction using Convolutional Auto-Encoders and Distance Fields

Curve reconstruction is a crucial task in various research domains, including Computer-Aided Design (CAD) and Reverse Engineering.Recovering a B-spline control polygon from a given set of points representing a curve remains an active area of study.We introduce a novel approach that leverages a distance field representation of the curve as input to two neural networks to reconstruct a closed B-spline.One network predicts distance fields to the control points, while the other estimates distances to the control polygon.Using these outputs, we determine the connectivity between control points, enabling the reconstruction of the B-spline control polygon.Our method is evaluated against state-of-the-art machine learning techniques and traditional optimization-based approaches.