Architectural Sketch to 3D Model: An Experiment on Simple-Form Houses

Abstract Transforming sketches into digital 3D models has been an enduring practice in the design process since first turn architecture. However, modeling is time-consuming, and software usually a cumbersome interface. Aiming to bridge gap between sketch model, we propose framework that can hand-drawn 2D mesh. The user draw on canvas through web-based interface, corresponding model will automatically be generated shown aside. downloaded or synchronized Rhino directly Grasshopper. proposed uses machine learning-based approach generate mesh from single by deforming template shape. Since have uneven surface, apply optimization step refine form, creating more usable architectural with planar faces sharper edges. We create Simple House Dataset, which consists of 5000 single-volume houses, train neural network. defined five categories house typologies - distinguished roof shape other geometric properties 1000 for each class parameters chosen randomly. Each includes 20 perspective line drawings different angles. Although limitation generalization ability makes it unlikely replace conventional today, fast transformation allows architects explore possibility various forms may speed up early stage. code this project dataset published Github [29].