Research map / Video / Video-language understanding
Compositional Physical Reasoning of Objects and Events From Videos
Zhenfang Chen, Shilong Dong, Kexin Yi, Yunzhu Li, Mingyu Ding, Antonio B. Torralba, Joshua B. Tenenbaum, Chuang Gan
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 reference videos for property learning (ablation) | PCR · ComPhy (validation set) | factual accuracy: 68.7 -> 78.3 | Improved |
| add reference videos for property learning (ablation) | PCR · ComPhy (validation set) | predictive accuracy (per question): 28.1 -> 56.5 | Improved |
| add reference videos for property learning (ablation) | PCR · ComPhy (validation set) concept learning | charge classification accuracy: 45.0 -> 68.1 | Improved |
| add counterfactual imagination (ablation) | PCR · ComPhy (validation set) | counterfactual accuracy (per question): 28.0 -> 50.2 | Improved |
| add counterfactual imagination (ablation) | PCR · ComPhy (validation set) | factual accuracy: 70.3 -> 78.3 | Improved |
| add counterfactual imagination (ablation) | PCR · ComPhy (validation set) concept learning | charge classification accuracy: 44.8 -> 68.1 | Improved |
| replace CPL with PCR (no dense property labels) | CPL · ComPhy (test set) | predictive accuracy (per question): 56.4 -> 62.0 | Improved |
| replace CPL with PCR (no dense property labels) | CPL · ComPhy (test set) | factual accuracy: 80.5 -> 76.0 | Hurt |
| add reference videos as input to ALOE | ALOE · ComPhy | factual accuracy: 54.3 -> 57.7 | Improved |
| add reference videos as input to MAC | MAC · ComPhy | factual accuracy: 68.6 -> 65.8 | Hurt |
| use large vision-language model GPT-4V instead of ALOE | ALOE · ComPhy | counterfactual accuracy (per question): 20.8 -> 8.9 | Hurt |
| use PCR instead of MAC | MAC · ComPhy-DIV | factual accuracy: 51.7 -> 68.4 | Improved |
Closest papers by meaning
- Physical Property Understanding from Language-Embedded Feature Fields (CVPR 2024 · similarity 0.88)
- ContPhy: Continuum Physical Concept Learning and Reasoning from Videos (ICML 2024 · similarity 0.88)
- Compositional 4D Dynamic Scenes Understanding with Physics Priors for Video Question Answering (ICLR 2025 · similarity 0.86)
- Beyond Static Vision: Scene Dynamic Field Unlocks Intuitive Physics Understanding in Multi-modal Large Language Models (ICLR 2026 · similarity 0.86)
- QUANTIPHY: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models (CVPR 2026 · similarity 0.85)
- MAPS: Advancing Multi-Modal Reasoning in Expert-Level Physical Science (ICLR 2025 · similarity 0.85)
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.
