Sketch-Based Modeling of Parametric Shapes

Abstract

We demonstrate a sketch-based modeling system running on a multi-touch pen tablet. Our system takes inspiration from the work of Nishida et al. [NGDGA*16], who proposed to use deep convolutional networks to interpret sketches of parametric shapes. While Nishida et al. applied this approach to the creation of procedural buildings, we focus on the creation of simple shapes (cuboids, cylinders, cones, spheres, pyramids) that users can assemble to create more complex objects. In this poster we describe the main components of our system - the deep convolutional networks used for sketch interpretation, the training data, the user interface, and the overall software architecture that combines these components. We will allow conference attendees to test our system on a pen tablet.