Research map / Video / Video-language understanding
From Evaluation to Enhancement: Benchmarking and Improving Think-with-Video Reasoning for Video Generative Models
Meng Luo, Yicheng Liu, Jiahao Wang, Yuanxing Zhang, Xin Tao, Pengfei Wan, Kun Gai, Hao Fei
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 Vid-PRE prompt rewriter | Wan2.2 · MME-CoF | overall score: 1.72 -> 2.58 | Improved |
| add Vid-PRE prompt rewriter | Wan2.2 · V-ReasonBench | average accuracy: 26.65 -> 44.48 | Improved |
| add Vid-PRE prompt rewriter | Wan2.2 · V-ReasonBench spatial cognition | accuracy: 1.32 -> 36.25 | Improved |
| add Vid-PRE prompt rewriter | Wan2.2 · VWG-Bench | overall score (1-5): 2.13 -> 2.51 | Improved |
| add Vid-PRE prompt rewriter | Wan2.2 · VWG-Bench | implicit rule following: 1.56 -> 2.12 | Improved |
| use Vid-PRE instead of Wan native prompt extension | Wan2.5 · MME-CoF | overall score: 2.84 -> 3.07 | Improved |
| use Vid-PRE instead of Wan native prompt extension | Wan2.6-Flash · MME-CoF | overall score: 2.82 -> 3.09 | Improved |
| use Vid-PRE instead of Wan native prompt extension | Wan2.5 · V-ReasonBench | average accuracy: 41.06 -> 51.19 | Improved |
| use Vid-PRE instead of Wan native prompt extension | Wan2.6-Flash · V-ReasonBench | average accuracy: 39.50 -> 52.57 | Improved |
| use Vid-PRE instead of Wan native prompt extension | Wan2.6-Flash · VWG-Bench | overall score (1-5): 3.04 -> 3.26 | Improved |
| add GRPO stage after SFT of Vid-PRE (ablation) | Wan2.2 · V-ReasonBench | average accuracy: 40.26 -> 44.48 | Improved |
| add GRPO stage after SFT of Vid-PRE (ablation) | Wan2.6-Flash · V-ReasonBench | average accuracy: 45.96 -> 52.57 | Improved |
Closest papers by meaning
- CollabVR: Collaborative Video Reasoning with Vision-Language and Video Generation Models (NeurIPS 2026 · similarity 0.93)
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.90)
- Think, then Score: Decoupled Reasoning and Scoring for Video Reward Modeling (NeurIPS 2026 · similarity 0.90)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.90)
- Video-Bench: Human-Aligned Video Generation Benchmark (CVPR 2025 · similarity 0.90)
- VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning (NeurIPS 2025 · similarity 0.89)
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.
