GeoFocus: Blending Efficient Global-to-Local Perception for Multimodal Geometry Problem-Solving
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866910016493060096 |
|---|---|
| author | Deng, Linger Liu, Yuliang Yu, Wenwen Zhang, Zujia Ju, Jianzhong Luo, Zhenbo Bai, Xiang |
| author_facet | Deng, Linger Liu, Yuliang Yu, Wenwen Zhang, Zujia Ju, Jianzhong Luo, Zhenbo Bai, Xiang |
| contents | Geometry problem-solving remains a significant challenge for Large Multimodal Models (LMMs), requiring not only global shape recognition but also attention to intricate local relationships related to geometric theory. To address this, we propose GeoFocus, a novel framework comprising two core modules. 1) Critical Local Perceptor, which automatically identifies and emphasizes critical local structure (e.g., angles, parallel lines, comparative distances) through thirteen theory-based perception templates, boosting critical local feature coverage by 61% compared to previous methods. 2) VertexLang, a compact topology formal language, encodes global figures through vertex coordinates and connectivity relations. By replacing bulky code-based encodings, VertexLang reduces global perception training time by 20% while improving topology recognition accuracy. When evaluated in Geo3K, GeoQA, and FormalGeo7K, GeoFocus achieves a 4.7% accuracy improvement over leading specialized models and demonstrates superior robustness in MATHVERSE under diverse visual conditions. Project Page -- https://github.com/dle666/GeoFocus |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_08524 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | GeoFocus: Blending Efficient Global-to-Local Perception for Multimodal Geometry Problem-Solving Deng, Linger Liu, Yuliang Yu, Wenwen Zhang, Zujia Ju, Jianzhong Luo, Zhenbo Bai, Xiang Computer Vision and Pattern Recognition Geometry problem-solving remains a significant challenge for Large Multimodal Models (LMMs), requiring not only global shape recognition but also attention to intricate local relationships related to geometric theory. To address this, we propose GeoFocus, a novel framework comprising two core modules. 1) Critical Local Perceptor, which automatically identifies and emphasizes critical local structure (e.g., angles, parallel lines, comparative distances) through thirteen theory-based perception templates, boosting critical local feature coverage by 61% compared to previous methods. 2) VertexLang, a compact topology formal language, encodes global figures through vertex coordinates and connectivity relations. By replacing bulky code-based encodings, VertexLang reduces global perception training time by 20% while improving topology recognition accuracy. When evaluated in Geo3K, GeoQA, and FormalGeo7K, GeoFocus achieves a 4.7% accuracy improvement over leading specialized models and demonstrates superior robustness in MATHVERSE under diverse visual conditions. Project Page -- https://github.com/dle666/GeoFocus |
| title | GeoFocus: Blending Efficient Global-to-Local Perception for Multimodal Geometry Problem-Solving |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2602.08524 |