Strokes2Surface: Recovering Curve Networks from 4D Architectural Design Sketches

Freeform architectural modelling often involves two basic steps: concept design and digital modelling. The First stage is usually conveyed via sketching, through which the design form eventually evolves. In the second stage, sketches serve as a visual reference for concept 3D CAD modelling while requiring a considerable amount of time editing it into the desired shape, which is utilised further for downstream engineering and fabrication applications. 

This research is aimed toward automated 3D digital modelling from concept design sketches (based on a 4D architectural design sketching application developed within our research group) through a novel pipeline composed of two modules: stroke beautification and surface mesh inference. The stroke beautification module relies mainly on two ML models, one being responsible for stroke type classification, while the other one is responsible for clustering 4D strokes into different groups, each representing a single curve which is thoroughly reconstructed and approximated by a cubic NURBS curve. Subsequently, surface mesh inference is responsible for forming a well-connected curve network from beautified strokes by recovering the desired topology, and triangulating its respective constituent simple cycles, resulting in the user-intended final shape. 

In this talk, the underlying mathematical and computational aspects of the aforementioned pipeline will be presented and also, some of the latest qualitative and quantitative evaluation results will be demonstrated.