Reproduce “A2Z-10M+: Geometric Deep Learning with A-to-Z BRep Annotations for AI-Assisted CAD Modeling and Reverse Engineering” 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
Pritham Kumar Jena, Bhavika Baburaj, Tushar Anand, Vedant Dutta, Vineeth Ulavala, Sk Aziz Ali
A2Z-10M+ is a comprehensive, multi-modal dataset from BITS Pilani, Hyderabad, featuring over 1 million CAD models with 10 million annotations, including high-resolution 3D scans, multi-level hand-drawn sketches, BRep information, and VLM-generated textual descriptions. It significantly improves performance for geometric deep learning tasks like BRep boundary and junction detection by providing high-quality, diverse training data.
Key insights
- Developing a large-scale, multi-modal dataset with high-fidelity BRep annotations, realistic scan simulations, and diverse sketches is crucial for advancing AI-assisted CAD modeling.
- Synthesizing realistic 3D scan data and artist-level 3D sketches, coupled with automated VLM-based textual descriptions, can bridge the gap between conceptual design inputs and precise BRep geometry.
- High-quality, comprehensive annotations on a vast dataset enable existing geometric deep learning models to achieve substantial performance improvements in fundamental CAD reverse engineering tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.