Research map / Video / Video generation
Physics Question Scene Graph: Fine-grained Evaluation of Physical Plausibility in Text-to-Video Generation
Atin Pothiraj, Jaemin Cho, Yue Zhang, Elias Stengel-Eskin, Mohit Bansal
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 logical dependency graph between questions (ablation) | PQSG · FinePhyEval overall | Pearson r (GPT-5 QA): 0.44 -> 0.48 | Improved |
| add logical dependency graph between questions (ablation) | PQSG · FinePhyEval overall | Pearson r (human QA): 0.75 -> 0.80 | Improved |
| add fine-grained question decomposition (ablation) | PQSG · FinePhyEval overall | Pearson r (GPT-5 QA): 0.40 -> 0.48 | Improved |
| add fine-grained question decomposition (ablation) | PQSG · FinePhyEval overall | Pearson r (human QA): 0.68 -> 0.80 | Improved |
| add reference video as input to question generation (ablation) | PQSG · FinePhyEval overall | Pearson r: 0.478 -> 0.418 | Hurt |
| add separate action vs physics prompt wording (ablation) | PQSG · FinePhyEval overall | Pearson r: 0.478 -> 0.407 | Hurt |
| use PQSG fine-grained questions instead of direct VQA | Gemini-2.5-Pro · FinePhyEval overall | Pearson r: 0.382 -> 0.467 | Improved |
| use PQSG instead of DSG question graph | Gemini-2.5-Pro · FinePhyEval overall | Pearson r: 0.302 -> 0.467 | Improved |
| use GPT-5.5 instead of Gemini-2.5-Pro as QA model | PQSG · FinePhyEval overall | Pearson r: 0.467 -> 0.478 | Improved |
| use GPT-5.5 instead of Gemini-2.5-Pro as QA model | PQSG · QA accuracy (physics) | accuracy (%): 61.5 -> 64.6 | Improved |
| use human QA instead of GPT-5.5 QA | PQSG · Physics-Physics per-category correlation | Pearson r: 0.48 -> 0.57 | Improved |
| add PQSG to VideoPhy-2-Autoeval | VideoPhy-2-Autoeval · VideoPhy-2 (SA) | Pearson r: 0.450 -> 0.550 | Improved |
Closest papers by meaning
- Towards World Simulator: Crafting Physical Commonsense-Based Benchmark for Video Generation (ICML 2025 · similarity 0.87)
- Davidsonian Scene Graph: Improving Reliability in Fine-grained Evaluation for Text-to-Image Generation (ICLR 2024 · similarity 0.87)
- Compositional 4D Dynamic Scenes Understanding with Physics Priors for Video Question Answering (ICLR 2025 · similarity 0.87)
- SOK-Bench: A Situated Video Reasoning Benchmark with Aligned Open-World Knowledge (CVPR 2024 · similarity 0.87)
- Q-Bench-Video: Benchmark the Video Quality Understanding of LMMs (CVPR 2025 · similarity 0.86)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 2025 · similarity 0.86)
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.
