Graph-CAD

Learning Hierarchical and Geometry-Aware Graph Representations for Text-to-CAD

Shengjie Gong, Wenjie Peng, Hongyuan Chen, Gangyu Zhang, Yunqing Hu, Huiyuan Zhang, Shuangping Huang, Tianshui Chen

Official Code Repository | Paper | Model Weights

---

Go To The Official Graph-CAD Repository

Overview

This repository is the laboratory entry point for Graph-CAD, a graph-mediated Text-to-CAD framework for long-horizon CAD code generation.
The actively maintained source code is hosted in the official project repository linked below.

Jump To The Official Resources

• Official code repository: https://github.com/EESJGong/Graph-CAD
• Paper: https://openreview.net/pdf?id=oKMomThD6n
• Model weights: https://modelscope.cn/models/JackeySmile/graph-cad

What Is Graph-CAD?

Graph-CAD introduces a hierarchical and geometry-aware intermediate graph for Text-to-CAD generation. Instead of directly decoding natural language into executable Blender bpy code, it decomposes the task into three stages:

1. predict a geometry-aware decomposition graph
2. convert the graph into CAD action sequences
3. generate executable Blender code

This design improves geometric fidelity, structural consistency, and geometric constraint satisfaction on complex CAD assemblies.

Note

• This repository is maintained as the lab-facing entry page.
• For the latest code, setup instructions, evaluation scripts, figures, and updates, please visit the official repository:

Citation

@inproceedings{gonglearning,
  title={Learning Hierarchical and Geometry-Aware Graph Representations for Text-to-CAD},
  author={Gong, Shengjie and Peng, Wenjie and Chen, Hongyuan and Zhang, Gangyu and Hu, Yunqing and Zhang, Huiyuan and Huang, Shuangping and Chen, Tianshui},
  booktitle={The Fourteenth International Conference on Learning Representations}
}