Research map / Vision-language and multimodal / Multimodal LLMs
MultiVerse: A Multi-Turn Conversation Benchmark for Evaluating Large Vision and Language Models
Young-Jun Lee, Byung-Kwan Lee, Jianshu Zhang, Yechan Hwang, Byungsoo Ko, Han-Gyu Kim, Dongyu Yao, Xuankun Rong, Eojin Joo, Seung-Ho Han, Bowon Ko, Ho-Jin Choi
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 |
|---|---|---|---|
| use ground-truth (Oracle) dialogue history instead of self-generated history | Qwen2.5-VL-72B · MultiVerse | gap vs Self-Prediction: -44.64 | Improved |
| use ground-truth (Oracle) dialogue history instead of self-generated history | Qwen2.5-VL-7B · MultiVerse | gap vs Self-Prediction: -30.44 | Improved |
| provide more interaction turns of golden history | InternVL2.5-2B · MultiVerse (Oracle) | checklist x quality score: 17.79 -> 34.32 | Improved |
| provide more interaction turns of golden history | LLaVA-1.5-7B · MultiVerse (Oracle) | checklist x quality score: 9.10 -> 31.81 | Improved |
| provide more interaction turns of golden history | Qwen2.5-VL-72B · MultiVerse (Oracle) | checklist x quality score: 52.05 -> 46.19 | Hurt |
| provide more interaction turns of golden history | Claude-3.5-Sonnet · MultiVerse (Oracle) | checklist x quality score: 46.60 -> 45.00 | No clear effect |
| provide more interaction turns of golden history | GPT-4o · MultiVerse (Oracle) | checklist x quality score: 48.56 -> 49.12 | No clear effect |
| scale Qwen2.5-VL from 7B to 72B | Qwen2.5-VL · MultiVerse (Oracle) | checklist x quality score: 48.15 -> 47.95 | No clear effect |
| scale Qwen2.5-VL from 7B to 72B | Qwen2.5-VL · MultiVerse (Oracle, goal: calculation) | checklist x quality score: 44.86 -> 47.15 | Improved |
| scale Qwen2.5-VL from 7B to 72B | Qwen2.5-VL · MultiVerse (Oracle, goal: creation) | checklist x quality score: 44.45 -> 39.82 | Hurt |
| scale InternVL2.5 from 1B to 4B | InternVL2.5 · MultiVerse (Oracle) | checklist x quality score: 21.22 -> 35.66 | Improved |
| scale LLaMA-3.2-Vision from 11B to 90B | LLaMA-3.2-Vision · MultiVerse (Oracle) | checklist x quality score: 21.61 -> 35.78 | Improved |
Closest papers by meaning
- ConvBench: A Multi-Turn Conversation Evaluation Benchmark with Hierarchical Ablation Capability for Large Vision-Language Models (NeurIPS 2024 · similarity 0.92)
- MMDU: A Multi-Turn Multi-Image Dialog Understanding Benchmark and Instruction-Tuning Dataset for LVLMs (NeurIPS 2024 · similarity 0.89)
- MMIE: Massive Multimodal Interleaved Comprehension Benchmark for Large Vision-Language Models (ICLR 2025 · similarity 0.88)
- DRESS: Instructing Large Vision-Language Models to Align and Interact with Humans via Natural Language Feedback (CVPR 2024 · similarity 0.88)
- Found in Conversation: LLMs Teach Themselves to Close the Multi-Turn Gap (NeurIPS 2026 · similarity 0.88)
- LLaVAShield: Safeguarding Multimodal Multi-Turn Dialogues in Vision-Language Models (CVPR 2026 · similarity 0.88)
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.
