GeoSym127K: Scalable Symbolically-verifiable Synthesis for Multimodal Geometric Reasoning
arXiv cs.CV·Jinhao Jing, Zheng Ma, Jinwei Liang, Qiannian Zhao, Shawn Chen, Jing Yang, Por Lip Yee, Prayag Tiwari, Jingjing Bai, Benyou Wang, Lewei Lu, Zhan Su
1d agoQuick Take
GeoSym127K introduces a scalable neuro-symbolic framework for enhanced geometric reasoning in multimodal models.
Key Points
- GeoSym Engine leverages type-conditional grammar for precise geometric reasoning.
- Dataset includes 51K images and 127K questions with symbolic ground truths.
- Qwen3-VL-8B model shows significant performance improvements on geometry tasks.
📖 Reader Mode
~2 min readAuthors:Jinhao Jing, Zheng Ma, Jinwei Liang, Qiannian Zhao, Shawn Chen, Jing Yang, Por Lip Yee, Prayag Tiwari, Jingjing Bai, Benyou Wang, Lewei Lu, Zhan Su
Abstract:Large Multimodal Models (LMMs) often struggle with geometric reasoning due to visual hallucinations and a lack of mathematically precise Chain-of-Thought (CoT) data. To address this, we propose the GeoSym Engine, an automated and scalable neuro-symbolic framework. By leveraging a type-conditional grammar and an analytic SymGT Solver, it derives exact symbolic ground truths and seamlessly integrates with a robust rendering pipeline to produce high-precision geometric diagrams. Using this engine, we construct GeoSym127K, a difficulty-stratified dataset featuring 51K high-resolution images, 127K questions with symbolic ground truths, and 55K answer-verified CoT QA pairs. We also introduce GeoSym-Bench, an expert-curated suite of 511 complex samples for rigorous evaluation. Through extensive supervised fine-tuning (SFT), we demonstrate that GeoSym drives concentrated improvements specifically on diagram-dependent and multi-step geometry tasks. Our Qwen3-VL-8B model gains an absolute +22.21% on the MathVerse Vision-Only subset and reaches 61.52% (+6.19% improvement) on WeMath, mitigating long-horizon logic fragmentation and outperforming advanced closed-source models like Doubao-1.8. Furthermore, applying Reinforcement Learning with Verifiable Rewards (RLVR) via GRPO reveals that initializing from structural SFT checkpoints substantially elevates the performance ceiling over zero-shot RL. Driven by deterministic exact-match signals, this showcases the robust scaling potential of our verifiable reasoning synthesis. Datasets and code are available at this https URL and this https URL.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.16371 [cs.CV] |
| (or arXiv:2605.16371v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2605.16371 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jinhao Jing [view email]
[v1]
Sun, 10 May 2026 13:13:47 UTC (6,173 KB)
— Originally published at arxiv.org
More from arXiv cs.CV
See more →arXiv cs.CV·Yuiko Sakuma, Masakazu Yoshimura, Marcel Gr\"opl, Zitang Sun, Junji Otsuka, Atsushi Irie, Takeshi Ohashi
17h ago FeaturedOriginal
Structuring Open-Ended NAS: Semi-Automated Design Knowledge Structuring with LLMs for Efficient Neural Architecture Search
AI Summary
This paper presents FairNAD, a semi-automated approach for efficient neural architecture search using structured design knowledge.
