ExtrudeNet: Unsupervised Inverse Sketch-and-Extrude for Shape Parsing

Sketch-and-extrude is a common and intuitive modeling process in computeraided design. This paper studies the problem of learning the shape given in theform of point clouds by inverse sketch-and-extrude. We present ExtrudeNet, anunsupervised end-to-end network for discovering sketch and extrude from pointclouds. Behind ExtrudeNet are two new technical components: 1) an effectiverepresentation for sketch and extrude, which can model extrusion with freeformsketches and conventional cylinder and box primitives as well; and 2) anumerical method for computing the signed distance field which is used in thenetwork learning. This is the first attempt that uses machine learning toreverse engineer the sketch-and-extrude modeling process of a shape in anunsupervised fashion. ExtrudeNet not only outputs a compact, editable andinterpretable representation of the shape that can be seamlessly integratedinto modern CAD software, but also aligns with the standard CAD modelingprocess facilitating various editing applications, which distinguishes our workfrom existing shape parsing research. Code is released athttps://github.com/kimren227/ExtrudeNet.