Research map / Video / Video generation
VideoPhy-2: A Challenging Action-Centric Physical Commonsense Evaluation in Video Generation
Hritik Bansal, Clark Peng, Yonatan Bitton, Roman Goldenberg, Aditya Grover, Kai-Wei Chang
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 |
|---|---|---|---|
| evaluate on the hard subset instead of the full dataset (ablation) | Wan2.1-14B · VideoPhy-2 | joint performance (%): 32.6 -> 21.9 | Hurt |
| closed-source model (Ray2, Sora) vs open models | video generators · VideoPhy-2 | joint performance (%): Ray2 20.3 vs Wan2.1-14B 32.6 | No clear effect |
| sports/physical activities vs object interactions prompts | video generators · VideoPhy-2 | joint performance (%): Wan2.1-14B PA 31.5 vs OI 36.2 | Hurt |
| optimize aesthetics and motion quality | video generators · VideoPhy-2 | correlation with physical commonsense: PC vs aesthetics 0.09, vs motion 0.002 | No clear effect |
| fine-tune VideoCon-Physics on 50K human annotations (multi-task) to get AutoEval | VideoPhy-2-AutoEval · unseen prompts | Pearson correlation avg (x100): 28.5 -> 42.0 | Improved |
| fine-tuned AutoEval vs Gemini-2.0-Flash on physical commonsense | VideoPhy-2-AutoEval · unseen prompts | PC correlation (x100): 11.0 -> 37.0 | Improved |
| fine-tuned AutoEval on physical rule classification vs Gemini-2.0-Flash | VideoPhy-2-AutoEval · unseen prompts | accuracy: 59.2 -> 78.7 | Improved |
| fine-tuned AutoEval on joint score vs VideoCon-Physics | VideoPhy-2-AutoEval · unseen prompts | F1: 2.6 -> 51.1 | Improved |
Closest papers by meaning
- VideoPhy: Evaluating Physical Commonsense for Video Generation (ICLR 2025 · similarity 0.94)
- Towards World Simulator: Crafting Physical Commonsense-Based Benchmark for Video Generation (ICML 2025 · similarity 0.90)
- Video-Bench: Human-Aligned Video Generation Benchmark (CVPR 2025 · similarity 0.89)
- VBench: Comprehensive Benchmark Suite for Video Generative Models (CVPR 2024 · similarity 0.87)
- Stable Cinemetrics : Structured Taxonomy and Evaluation for Professional Video Generation (NeurIPS 2025 · similarity 0.87)
- Needle In A Video Haystack: A Scalable Synthetic Evaluator for Video MLLMs (ICLR 2025 · 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.
