Research map / Video / Video-language understanding
ComPhy: Compositional Physical Reasoning of Objects and Events from Videos
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 four reference videos for inferring physical properties (ablation) | Target-video-only GNN · ComPhy physical-property labels, mass | accuracy: 59.4% -> 90.4% | Helped |
| Add four reference videos for inferring physical properties (ablation) | Target-video-only GNN · ComPhy physical-property labels, charge edge | accuracy: 70.2% -> 90.8% | Helped |
| Concatenate reference-video features (ablation) | CNN-LSTM · ComPhy test, factual | accuracy: 46.6 -> 41.9 | Hurt |
| Concatenate reference-video features (ablation) | MAC · ComPhy test, factual | accuracy: 68.6 -> 65.8 | Hurt |
| Concatenate reference-video features (ablation) | ALOE · ComPhy test, predictive per option | accuracy: 65.9 -> 67.9 | Helped |
| Concatenate reference-video features (ablation) | ALOE · ComPhy test, counterfactual per question | accuracy: 20.8 -> 22.2 | Helped |
| Concatenate reference-video features (ablation) | CNN-LSTM · ComPhy test, predictive per option | accuracy: 59.5 -> 59.6 | No clear effect |
| Concatenate reference-video features (ablation) | MAC · ComPhy test, predictive per option | accuracy: 60.2 -> 60.2 | No clear effect |
| Explicitly model mass on graph nodes and relative charge on edges | CPL versus NS-DR+ · ComPhy predictive and counterfactual reasoning | accuracy | Helped |
Closest papers by meaning
- A Compositional Object-Based Approach to Learning Physical Dynamics (ICLR 2017 · similarity 0.89)
- Physical Property Understanding from Language-Embedded Feature Fields (CVPR 2024 · similarity 0.87)
- Beyond Static Vision: Scene Dynamic Field Unlocks Intuitive Physics Understanding in Multi-modal Large Language Models (ICLR 2026 · similarity 0.87)
- ContPhy: Continuum Physical Concept Learning and Reasoning from Videos (ICML 2024 · similarity 0.87)
- QUANTIPHY: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models (CVPR 2026 · similarity 0.86)
- VideoPhy: Evaluating Physical Commonsense for Video Generation (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.