Neural network model optimisation for 3D freeform surfaces from sketched curves

The induction of a NURBS freeform surface from an on-line sketch is presented in this paper. This work supports the inference of 3D geometry based on input 2D freehand sketches to support conceptual design. A multi-layer Perceptron (MLP) artificial neural network (ANN) was designed by a series of experiments and implemented to learn the relationship between input sketches and their expected 3D geometry. Experimentation was used to determine the optimal parameters and network architecture. The input data took the format of four constrained boundary strokes and after the ANN had inferred the 3D shape, a complete NURBS serface was generated using an interpolation algorithm