Hilbert-Geo: Solving Solid Geometric Problems
by Neural-Symbolic Reasoning

News

2026.02:  🎉🎉 Our paper "Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning" was accepted by CVPR2026. The dataset link：https://huggingface.co/datasets/WendingGao/Hilbert-Geo2K

Structure

SolidGeoSolver/
├── api/
│   ├── base.py
│   ├── claude_api.py
├── core/
│   ├── fgps/
│   ├── gdl/
│   ├── files/
│   └── hilbert_geo/
└── data/
    └── hilbert_geo_v2/

• api/ contains prompt-and-call helpers for model APIs.
• core/hilbert_geo/ is the package.
• core/gdl/ and core/files/t_info.json keep the predicate library, theorem bank, and theorem metadata with the core code.
• data/hilbert_geo_v2/ contains sample subset for repository display and quick testing. https://github.com/CHYYYYYYYY/SolidGeoSolver/tree/main/data/hilbert_geo7k_v2

Quick Start

Install dependencies, you may need to manually download hilbertgeo in requirement.txt at present:

pip install -r requirements.txt

Run the interactive solver:

python core/fgps/run.py --func run

Run search:

python core/fgps/search.py --func search --method fw --strategy bfs

By default:

• datasets are loaded from data/
• logs are written to core/fgps/
• GDL and theorem metadata are loaded from core/ when they are not present in the dataset folder

Notes

• The Python package name is now hilbert_geo.
• The repository display name used in docs is Hilbert-Geo.
• The sample dataset folder is renamed to hilbert_geo_v2 and only includes the first 1000 problems.

✍️ Citation

If you use our work and are inspired by our work, please consider cite us:

@InProceedings{Xu_2026_CVPR,
    author    = {Xu, Ruoran and Cheng, Haoyu and Dong, Bin and Wang, Qiufeng},
    title     = {Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2026},
    pages     = {9658-9667}
}