Research map / Video / Video-language understanding
Benchmarking Scientific Understanding and Reasoning for Video Generation using VideoScience-Bench
Lanxiang Hu, Abhilash Shankarampeta, Yixin Huang, Zilin Dai, Haoyang Yu, Yujie Zhao, Haoqiang Kang, Daniel Zhao, Tajana Rosing, Hao Zhang
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 checklist and key-frame evidence to the VLM judge (ablation) | VSci-Judge (GPT-5 pro) · VideoScience-Bench | Kendall tau vs human: 0.81 -> 0.90 | Improved |
| add CV evidence tools (ByteTrack, RAFT, etc.) on top of checklist (ablation) | VSci-Judge (CL) · VideoScience-Bench | Kendall tau vs human: 0.90 -> 0.90 | No clear effect |
| add CV evidence on top of checklist (ablation) | VSci-Judge (CL) · VideoScience-Bench | Immutability and Coherence overestimation gap: Immutability +0.24 -> +0.18; Coherence +0.26 -> +0.21 | Improved |
| add checklist and CV evidence (ablation) | VSci-Judge · VideoScience-Bench | Dynamism overestimation gap: +0.32 -> +0.29 | No clear effect |
| replace GPT-4o judge with reasoning model GPT-5 pro (ablation) | VSci-Judge · VideoScience-Bench | gap to human, Prompt Consistency: +0.223 -> -0.096 | Improved |
| replace GPT-4o judge with GPT-5 pro (ablation) | VSci-Judge · VideoScience-Bench | gap to human, Phenomenon Congruency: +0.125 -> -0.218 | Mixed |
| use VSci-Judge instead of PhyGenEval | video evaluation · VideoScience-Bench | Spearman rho vs human: 0.61 -> 0.89 | Improved |
| use VSci-Judge instead of VideoScore2 | video evaluation · VideoScience-Bench | Spearman rho vs human: 0.29 -> 0.89 | Improved |
Closest papers by meaning
- Apple-$\pi$: Benchmarking Thinking with Video Towards Law-Grounded Physical Intelligence (NeurIPS 2026 · similarity 0.90)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.89)
- VideoPhy: Evaluating Physical Commonsense for Video Generation (ICLR 2025 · similarity 0.88)
- Impossible Videos (ICML 2025 · similarity 0.88)
- Towards World Simulator: Crafting Physical Commonsense-Based Benchmark for Video Generation (ICML 2025 · similarity 0.88)
- WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (NeurIPS 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.
