Research map / Vision-language and multimodal / Multimodal LLMs
VisNumBench: Evaluating Number Sense of Multimodal Large Language Models
Tengjin Weng, Jingyi Wang, Wenhao Jiang, Zhong Ming
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 |
|---|---|---|---|
| fine-tune InternVL2-8B with MMPR preference data (MPO) (ablation) | InternVL2-8B · VisNumBench-Synthetic | accuracy (%): 39.56 -> 40.65 | Improved |
| fine-tune InternVL2-8B with MMPR preference data (MPO) (ablation) | InternVL2-8B · VisNumBench-Real | accuracy (%): 39.58 -> 39.91 | No clear effect |
| fine-tune LLaVA-v1.5-13B on MathV360K (Math-LLaVA) (ablation) | LLaVA-v1.5-13B · VisNumBench-Synthetic | accuracy (%): 32.15 -> 35.81 | Improved |
| fine-tune LLaVA-v1.5-13B on MathV360K (Math-LLaVA) (ablation) | LLaVA-v1.5-13B · VisNumBench-Real | accuracy (%): 40.02 -> 33.15 | Hurt |
| add chain-of-thought training (LLaVA-o1-100k) (ablation) | Llama-VL-3.2-11B · VisNumBench-Synthetic | accuracy (%): 43.92 -> 45.50 | Improved |
| add chain-of-thought training (LLaVA-o1-100k) (ablation) | Llama-VL-3.2-11B · VisNumBench-Real | accuracy (%): 43.24 -> 38.36 | Hurt |
| add R1-style rule-based RL reasoning training (R1-Onevision) (ablation) | Qwen2.5-VL-7B · VisNumBench-Synthetic | accuracy (%): 46.19 -> 38.87 | Hurt |
| add R1-style rule-based RL reasoning training (R1-Onevision) (ablation) | Qwen2.5-VL-7B · VisNumBench-Real | accuracy (%): 41.02 -> 38.25 | Hurt |
| add CoT prompt (ablation) | Gemini 2.0 Flash · VisNumBench-Synthetic | accuracy (%): 57.57 -> 60.14 | Improved |
| add CoT prompt (ablation) | Qwen2.5-VL-72B · VisNumBench-Synthetic | accuracy (%): 58.46 -> 56.68 | Hurt |
| add CoT prompt (ablation) | InternVL2.5-78B · VisNumBench-Synthetic | accuracy (%): 56.18 -> 53.21 | Hurt |
| add CoT prompt (ablation) | Qwen2.5-VL-72B · VisNumBench-Real | accuracy (%): 53.33 -> 51.11 | Hurt |
Closest papers by meaning
- Benchmarking Multimodal Mathematical Reasoning with Explicit Visual Dependency (NeurIPS 2026 · similarity 0.90)
- MV-MATH: Evaluating Multimodal Math Reasoning in Multi-Visual Contexts (CVPR 2025 · similarity 0.89)
- VisuLogic: A Benchmark for Evaluating Visual Reasoning in Multi-modal Large Language Models (ICLR 2026 · similarity 0.88)
- MathSE: Improving Multimodal Mathematical Reasoning via Self-Evolving Iterative Reflection and Reward-Guided Fine-Tuning (AAAI 2026 · similarity 0.87)
- DynaMath: A Dynamic Visual Benchmark for Evaluating Mathematical Reasoning Robustness of Vision Language Models (ICLR 2025 · similarity 0.87)
- VP-Bench: A Comprehensive Benchmark for Visual Prompting in Multimodal Large Language Models (AAAI 2026 · similarity 0.87)
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.
