BRepGround

BRepGround grounds a text query to faces and edges of a CAD B-Rep.

Paper · Code · Dataset

Usage

Install the futurecad environment from the code repository. Run from that repository's root:

hf download jhlee11/BRepGround --local-dir weights/brepground
python -m grounding_brep.serve \
  --config weights/brepground/config.yaml \
  --checkpoint weights/brepground/model.safetensors

To evaluate the model, prepare the STEP dataset's graph cache and run bash scripts/evaluate_grounding.sh. See dataset usage for the label-reading convention and commands.

The model uses BERT-large-uncased as its text encoder. Its tokenizer and configuration are included in bert-large-uncased/.

Citation

@misc{li2026highfidelitycadgenerationllmdriven,
  title={Towards High-Fidelity CAD Generation via LLM-Driven Program Generation and Text-Based B-Rep Primitive Grounding},
  author={Jiahao Li and Qingwang Zhang and Qiuyu Chen and Guozhan Qiu and Yunzhong Lou and Xiangdong Zhou},
  year={2026},
  eprint={2603.11831},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2603.11831},
}