Show HN: Shape Foundation Model (masked-token pretraining on CAD meshes)
Category: ai-ml
Tags: 3d-geometry, foundation-model, self-supervised-learning, cad-analysis, computer-vision
Score: 7.8/10 (Innovation: 8, Technical: 8, Documentation: 9, Utility: 6)
A 3D geometry foundation model that applies masked-token pretraining (like BERT) to CAD meshes, producing dense geometric embeddings for industrial CAD analysis. It's interesting because it adapts NLP-style self-supervised learning to 3D geometry with a novel MAGNO encoder and offers explainable predictions via per-token attribution.
Target audience: data engineers, ai researchers, cad engineers, computer vision engineers
Repository: https://github.com/simd-ai/shape-v2 · Python · Apache-2.0 · 1 stars
View on Hacker News