Research map / Vision-language and multimodal / Multimodal LLMs
Video-Holmes: Can MLLM Think like Holmes for Complex Video Reasoning?
JUNHAO CHENG, Yuying Ge, Teng Wang, Yixiao Ge, Jing Liao, Ying Shan
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 RL post-training for thinking | Gemini-2.0-Flash · Video-Holmes | accuracy: 30.6 -> 43.1 | Improved |
| add RL post-training for thinking | Gemini-2.0-Flash · MMVU | accuracy: 65.2 -> 64.5 | No clear effect |
| add RL post-training for thinking | Gemini-2.0-Flash · Video-MME | accuracy: 68.3 -> 69.4 | No clear effect |
| add RL video reasoning training (Video-R1) | Qwen2.5-VL-7B · Video-Holmes | overall accuracy: 27.8 -> 36.5 | Improved |
| add RL training (SEED-Bench-R1) | Qwen2.5-VL-7B · Video-Holmes | overall accuracy: 27.8 -> 33.5 | Improved |
| increase input frames from 32 to 80 (ablation) | Qwen2.5-VL-7B · Video-Holmes | accuracy: 27.8 -> 33.0 | Improved |
| increase input frames from 32 to 50 (ablation) | GPT-4o · Video-Holmes | accuracy: 42.0 -> 44.6 | Improved |
| add audio input (ablation) | Gemini-2.5-Pro · Video-Holmes | overall accuracy: 45.0 -> 51.3 | Improved |
| add audio input (ablation) | Qwen2.5-Omni-7B · Video-Holmes | overall accuracy: 16.4 -> 24.4 | Improved |
| enable CoT reasoning (ablation) | Video-R1 · Video-Holmes | accuracy: 28.2 -> 36.5 | Improved |
| enable CoT reasoning (ablation) | Qwen2.5-VL-7B · Video-Holmes | accuracy: 29.4 -> 27.8 | Hurt |
| enable CoT reasoning (ablation) | Gemini-2.0-Flash · Video-Holmes | accuracy: 28.5 -> 30.6 | Improved |
Closest papers by meaning
- VideoReasonBench: Can MLLMs Perform Vision-Centric Complex Video Reasoning? (ICLR 2026 · similarity 0.89)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.89)
- MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos (ICLR 2026 · similarity 0.88)
- Black Swan: Abductive and Defeasible Video Reasoning in Unpredictable Events (CVPR 2025 · similarity 0.88)
- HumanPCR: Probing MLLM Capabilities in Diverse Human-Centric Scenes (ICLR 2026 · similarity 0.88)
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · 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.
