Research map / Vision-language and multimodal / Multimodal LLMs
SimpleVQA: Multimodal Factuality Evaluation for Multimodal Large Language Models
Xianfu Cheng, Wei Zhang, Shiwei Zhang, Jian Yang, Xiangyuan Guan, Xianjie Wu, Xiang Li, Ge Zhang, Jiaheng Liu, Yuying Mai, Yutao Zeng, Zhoufutu Wen and 8 more
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 |
|---|---|---|---|
| add the atomic fact to the question prompt (ablation) | GPT-4o · SimpleVQA CFQ subset | accuracy (CO): 56.24% -> 61.69% | Improved |
| add the atomic fact to the question prompt (ablation) | Qwen2.5-VL-72B-Instruct · SimpleVQA CFQ subset | accuracy (CO): 51.67% -> 64.15% | Improved |
| add the atomic fact to the question prompt (ablation) | InternVL2.5-78B-MPO · SimpleVQA CFQ subset | accuracy (CO): 55.36% -> 69.95% | Improved |
| ask the atomic recognition question instead of the original complex question (ablation) | Qwen2.5-VL-72B-Instruct · SimpleVQA CFQ subset | accuracy (CO): 51.67% -> 59.58% | Improved |
| ask the atomic question instead of the original question (ablation) | GPT-4o · SimpleVQA CFQ subset | accuracy (CO): 56.24% -> 56.94% | No clear effect |
| ask the atomic question instead of the original question (ablation) | InternVL2.5-78B-MPO · SimpleVQA CFQ subset | accuracy (CO): 55.36% -> 55.71% | No clear effect |
| remove the image input (text-only question) (ablation) | GPT-4o · SimpleVQA | accuracy (CO): 47.2 -> 5.23% | Hurt |
| remove the image input (text-only question) (ablation) | Claude-3.5-Sonnet · SimpleVQA | accuracy (CO): 48.5 -> 2.02% | Hurt |
| evaluate on English questions instead of Chinese (ablation) | Doubao-vision-pro-128k · SimpleVQA | F-score: 53.6 -> 33.9 | Hurt |
| evaluate on English questions instead of Chinese (ablation) | Claude-3.5-Sonnet · SimpleVQA | F-score: 49.3 -> 52.8 | Mixed |
| scale InternVL2.5 from 8B-MPO to 78B-MPO (ablation) | InternVL2.5-MPO · SimpleVQA | F-score: 34.9 -> 46.8 | Improved |
| upgrade Qwen2.5-VL from 7B to 72B (ablation) | Qwen2.5-VL · SimpleVQA | F-score: 44.3 -> 50.8 | Improved |
Closest papers by meaning
- CCFQA: A Benchmark for Cross-Lingual and Cross-Modal Speech and Text Factuality Evaluation (AAAI 2026 · similarity 0.85)
- Multimodal Fact-Level Attribution for Verifiable Reasoning (ICML 2026 · similarity 0.84)
- Fine-Tuning Language Models for Factuality (ICLR 2024 · similarity 0.84)
- Toward Robust Hyper-Detailed Image Captioning: A Multiagent Approach and Dual Evaluation Metrics for Factuality and Coverage (ICML 2025 · similarity 0.83)
- Long-form factuality in large language models (NeurIPS 2024 · similarity 0.83)
- SEED-Bench: Benchmarking Multimodal Large Language Models (CVPR 2024 · similarity 0.82)
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.
