Point2CAD reconstructs complex CAD models from 3D point clouds. A point cloud is segmented into clusters corresponding to CAD faces. Each face is fitted with a geometric primitive or a parametric surface using a novel neural representation. Due to the analytic representation, the surfaces can be extended and intersected such that topology emerges, which is then used to clip the surface primitives.

Computer-Aided Design (CAD) model reconstruction from point clouds is an important problem at the intersection of computer vision, graphics, and machine learning; it saves the designer a significant amount of time when iterating on in-the-wild objects. Recent advancements in this direction achieve rather reliable semantic segmentation but still struggle to produce an adequate topology of the CAD model. In this work, we analyze the current state of the art for that ill-posed task and identify shortcomings of existing methods. We propose a hybrid analytic-neural reconstruction scheme that bridges the gap between segmented point clouds and structured CAD models and can be readily combined with different segmentation backbones. Moreover, to power the surface fitting stage, we propose a novel implicit neural representation of freeform surfaces, driving up the performance of our overall CAD reconstruction scheme. We extensively evaluate our method on the popular ABC benchmark of CAD models and set a new state-of-the-art for that dataset.

The window below provides a split view of CAD model input point clouds and their reconstructions with various methods, including flavors of Point2CAD.

Use the mouse or gestures to zoom, spin the model, and slide the splitter for a better view of reconstruction quality.

@misc{liu2023point2cad,
  author = {Liu, Yujia and Obukhov, Anton and Wegner, Jan Dirk and Schindler, Konrad},
  title = {Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds},
  year = {2023},
}