{"slug": "so-3-equivariant-learning-on-cad-boundary-representations", "title": "SO(3)-Equivariant Learning on CAD Boundary Representations", "summary": "Researchers at Ghent University's CVAMO group released SO3brep, an SO(3)-equivariant graph neural network that operates on CAD boundary-representation face-adjacency graphs, detailed in arXiv paper 2611.05449v1. SO3brep computes node features in local frames and maps them into a shared frame using Wigner D-matrices, coupling them with inter-face directions via Clebsch-Gordan tensor products so relative orientation is preserved through message passing and outputs stay SE(3)-invariant. Across three face-segmentation and two classification benchmarks with randomly rotated test parts, SO3brep posted the highest accuracy among compared B-rep and equivariant point cloud networks with the fewest parameters, with its largest gains in the few-labeled-parts setting typical of industrial CAD; code is at github.com/UGent-CVAMO/so3brep.", "body_md": "arXiv:2611.05449v1 Announce Type: new \nAbstract: Learning on boundary representations (B-reps) requires robustness to the pose of a part, since CAD models are not stored in a canonical orientation. Existing B-rep networks either operate on world-frame coordinates and normals, and thus degrade under rotation, or encode each face in a local reference frame. Local frames yield invariance, but reduce inter-face messages to scalars and discard the relative orientation of adjacent faces. We propose SO3brep, an SO(3)-equivariant graph neural network on the B-rep face-adjacency graph. Node features are computed in local frames and mapped into a shared frame by Wigner D-matrices. Clebsch-Gordan tensor products couple these features with the directions between adjacent faces, so that relative orientation is preserved throughout message passing and the output remains SE(3)-invariant. Across three face-segmentation and two classification benchmarks with randomly rotated test parts, SO3brep achieves the highest accuracy among the compared B-rep networks and equivariant point cloud networks, with the fewest parameters. Its advantage is largest with few labeled parts, a setting typical of industrial CAD. Code is available at https://github.com/UGent-CVAMO/so3brep.", "url": "https://wpnews.pro/news/so-3-equivariant-learning-on-cad-boundary-representations", "canonical_source": "https://arxiv.org/abs/2611.05449", "published_at": "2026-11-05 05:00:00+00:00", "updated_at": "2026-11-05 05:23:16.978767+00:00", "lang": "en", "topics": ["machine-learning", "ai-research", "neural-networks", "artificial-intelligence"], "entities": ["SO3brep", "Ghent University", "CVAMO", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/so-3-equivariant-learning-on-cad-boundary-representations", "markdown": "https://wpnews.pro/news/so-3-equivariant-learning-on-cad-boundary-representations.md", "text": "https://wpnews.pro/news/so-3-equivariant-learning-on-cad-boundary-representations.txt", "jsonld": "https://wpnews.pro/news/so-3-equivariant-learning-on-cad-boundary-representations.jsonld"}}