Reproduce “ComplexGen: CAD Reconstruction by B-Rep Chain Complex Generation” 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
Haoxiang Guo, Shilin Liu, Hao Pan, Yang Liu, Xin Tong, Baining Guo
ComplexGen reconstructs connected boundary-representation CAD models from point clouds by jointly predicting corners, curves, patches, and their adjacencies, then using global optimization and geometric refinement to resolve the model. Its face–edge topology error falls from 0.229 for separate detectors to 0.145. Patch-to-patch error is 0.191 versus HPNet’s 0.343, though HPNet has lower patch residual.
Key insights
- Modeling CAD reconstruction as a B-Rep chain complex enables the simultaneous and holistic generation of all B-Rep elements and their topology, inherently incorporating structural constraints.
- A deep neural network designed with cross-communication between different element groups (vertices, edges, faces) can effectively predict a coherent probabilistic B-Rep structure.
- Combining probabilistic neural predictions with a constrained global optimization step is effective in recovering definitive, topologically valid CAD models, integrating learning-based and traditional CAD principles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.