Reproduce “Point2Primitive: CAD Reconstruction from Point Cloud by Direct Primitive Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xinzhu Ma, Cheng Wang, Chen Tang, Bin Wang, Shixiang Tang, Yuan Meng, Yunhong Wang, Di Huang
Researchers from Beihang University, Shanghai AI Lab, and collaborators developed Point2Primitive, a framework that directly reconstructs editable CAD models from point clouds by predicting explicit parametric sketch primitives. This method achieves higher geometric fidelity and accuracy compared to existing approaches, producing models with sharp edges suitable for standard CAD software.
Key insights
- Directly predicting explicit parametric sketch primitives addresses the precision and editability limitations of implicit neural representations in CAD reconstruction.
- Decomposing the reconstruction into extrusion segmentation for topology and direct sketch primitive prediction for geometry simplifies the complex reverse engineering task.
- An improved transformer decoder can effectively perform precise parametric regression for geometric primitives by integrating geometric priors as explicit positional queries and refining parameters autoregressively.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.