CAD-SIGNet: CAD Language Inference from Point Clouds using Layer-wise Sketch Instance Guided Attention

Reverse engineering in the realm of Computer-Aided Design (CAD) has been alongstanding aspiration, though not yet entirely realized. Its primary aim isto uncover the CAD process behind a physical object given its 3D scan. Wepropose CAD-SIGNet, an end-to-end trainable and auto-regressive architecture torecover the design history of a CAD model represented as a sequence ofsketch-and-extrusion from an input point cloud. Our model learnsvisual-language representations by layer-wise cross-attention between pointcloud and CAD language embedding. In particular, a new Sketch instance GuidedAttention (SGA) module is proposed in order to reconstruct the fine-graineddetails of the sketches. Thanks to its auto-regressive nature, CAD-SIGNet notonly reconstructs a unique full design history of the corresponding CAD modelgiven an input point cloud but also provides multiple plausible design choices.This allows for an interactive reverse engineering scenario by providingdesigners with multiple next-step choices along with the design process.Extensive experiments on publicly available CAD datasets showcase theeffectiveness of our approach against existing baseline models in two settings,namely, full design history recovery and conditional auto-completion from pointclouds.