Reproduce “CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language Models” 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
Vladislav Pyatov, Gleb Bobrovskikh, Saveliy Galochkin, Nikita Boldyrev, Oleg Voynov, Alexander Filippov, Gonzalo Ferrer, Peter Wonka and 1 more
CADFS introduces a large-scale multimodal parametric CAD dataset and a framework that enables vision-language models to generate complex CAD design histories using FeatureScript, covering 15 distinct modeling operations. The system, developed by Pyatov et al., demonstrates improved geometric accuracy (e.g., a 40% reduction in Chamfer Distance over Cadrille) and diversity in both text-conditioned and image-based reconstruction compared to previous methods.
Key insights
- FeatureScript provides a robust, expressive, and industry-aligned representation for generative CAD, allowing for a broader range of modeling operations and precise geometric entity referencing.
- High-fidelity, semantically aligned textual annotations significantly improve geometric accuracy and diversity in text-conditioned CAD generation by fostering better alignment between intent and design logic.
- Scale-aware generation, where bounding box dimensions are provided, and higher input image resolution are critical for enhancing the geometric accuracy and validity of complex CAD model reconstructions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.