README.md

metadata

task_categories:
  - image-text-to-text
tags:
  - geometry
  - mathematical-reasoning
  - multimodal

GeoFocus-test

Paper | GitHub

This repository contains the test and evaluation data for GeoFocus, a framework for multimodal geometry problem-solving. GeoFocus addresses the challenge of geometry reasoning by blending efficient global and local perception through two core modules:

1. Critical Local Perceptor: Automatically identifies and emphasizes critical local structures (e.g., angles, parallel lines, comparative distances) through thirteen theory-based perception templates.
2. VertexLang: A compact topology formal language that encodes global figures through vertex coordinates and connectivity relations, reducing global perception training time while improving topology recognition accuracy.

This dataset is used to evaluate models on benchmarks including Geo3K, GeoQA, and FormalGeo7K, and demonstrates superior robustness in MATHVERSE.

Related Datasets

The training data used by GeoFocus is available at the following links:

• Global_Perceptor_Data
• Local_Perceptor_Data

Citation

If you find this dataset or the GeoFocus framework useful for your research, please cite:

@article{geofocus2026,
  title={GeoFocus: Blending Efficient Global-to-Local Perception for Multimodal Geometry Problem-Solving},
  author={Dle et al.},
  journal={arXiv preprint arXiv:2602.08524},
  year={2026}
}