Research map / Vision-language and multimodal / Multimodal LLMs
MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?
Renrui Zhang, Dongzhi Jiang, Yichi Zhang, Haokun Lin, Ziyu Guo, Pengshuo Qiu, Aojun Zhou, Pan Lu, Kai-Wei Chang, Peng Gao, Hongsheng Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| remove the diagram input (Text Dominant to Text Only) (ablation) | Qwen-VL-Max · MathVerse testmini | CoT-E score: +5.1% | Improved |
| remove the diagram input (Text Dominant to Text Only) (ablation) | InternLM-XComposer2 · MathVerse testmini | CoT-E score: +5.6% | Improved |
| remove the diagram input (Text Dominant to Text Only) (ablation) | GPT-4V · MathVerse testmini | CoT-E score: 63.1 -> 60.3 | Hurt |
| move Essential Condition from text into diagram (Text-lite to Vision-dominant) (ablation) | GPT-4V · MathVerse testmini | CoT-E score: -5.8% | Hurt |
| move Essential Condition from text into diagram (Text-lite to Vision-dominant) (ablation) | InternLM-XComposer2 · MathVerse testmini | CoT-E score: -3.9% | Hurt |
| remove Implicit Property from text (Text-lite to Vision-intensive) (ablation) | most MLLMs · MathVerse testmini | CoT-E score | No clear effect |
| fine-tune LLaVA-1.5 on geometry-only data (G-LLaVA) | LLaVA-1.5 · MathVerse Functions/Solid/Analytic | CoT-E score: -3.5% Anal, -5.6% Solid, -5.6% Functions | Mixed |
| use CoT evaluation instead of final-answer accuracy | GPT-4V · MathVerse testmini | score: +16.1% | Improved |
| math-specific training data in the MLLM | SPHINX-MoE and InternLM-XComposer2 · MathVerse testmini | CoT-E score: 25.9 and 22.8 vs 12.7 for LLaVA-1.5 | Improved |
Closest papers by meaning
- MV-MATH: Evaluating Multimodal Math Reasoning in Multi-Visual Contexts (CVPR 2025 · similarity 0.93)
- DynaMath: A Dynamic Visual Benchmark for Evaluating Mathematical Reasoning Robustness of Vision Language Models (ICLR 2025 · similarity 0.92)
- Math Blind: Failures in Diagram Understanding Undermine Reasoning in MLLMs (ICLR 2026 · similarity 0.92)
- Benchmarking Multimodal Mathematical Reasoning with Explicit Visual Dependency (NeurIPS 2026 · similarity 0.91)
- MINT-CoT: Enabling Interleaved Visual Tokens in Mathematical Chain-of-Thought Reasoning (NeurIPS 2025 · similarity 0.91)
- VisioMath: Benchmarking Figure-based Mathematical Reasoning in LMMs (ICLR 2026 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
