Research map / 3D vision / 3D generation
Newtonian Scene Understanding: Unfolding the Dynamics of Objects in Static Images
Check 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 |
|---|---|---|---|
| Use an intermediate Newtonian-scenario representation instead of direct trajectory regression | N³ · VIND test, average across 12 Newtonian scenarios | F-measure: 30.31 -> 55.96 | Helped |
| Use an intermediate Newtonian-scenario representation instead of direct regression followed by nearest-curve matching | N³ · VIND test, average across 12 Newtonian scenarios | F-measure: 37.05 -> 55.96 | Helped |
| Use N³ rather than direct trajectory regression | N³ · VIND test | Modified Hausdorff Distance: 7.32 -> 5.59 | Helped |
| Use N³ rather than Predictive-CNN for single-image flow estimation | N³ · VIND test | Angular Err.: 1.53 -> 1.29 | Helped |
| Use N³ rather than direct regression when evaluating on scene types held out from training | N³ · Unseen scene types, average across 12 Newtonian scenarios | F-measure: 25.76 -> 36.40 | Helped |
| Add the motion row / motion abstraction (ablation) | N³ · VIND test, average across 12 Newtonian scenarios | F-measure: 52.67 -> 55.96 | Helped |
| Add state-level supervision (ablation) | N³ · VIND test, average across 12 Newtonian scenarios | F-measure: 55.96 -> 56.10 | No clear effect |
| Add the motion row / motion abstraction (ablation) | N³ · VIND classification, 66 scenario-view classes | Avg. Accuracy: 20.37 -> 21.71 | Helped |
| Add state-level supervision while omitting the motion row (ablation) | N³-NV · VIND classification, 66 scenario-view classes | Avg. Accuracy: 20.37 -> 19.32 | Hurt |
| Add state-level supervision (ablation) | N³ · VIND classification, 66 scenario-view classes | Avg. Accuracy: 21.71 -> 21.94 | No clear effect |
Closest papers by meaning
- Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systems (ICML 2026 · similarity 0.87)
- Dynamics: Language-Based Representation for Inferring Rigid-Body Dynamics From Videos (CVPR 2026 · similarity 0.86)
- NewtonGen: Physics-consistent and Controllable Text-to-Video Generation via Neural Newtonian Dynamics (ICLR 2026 · similarity 0.86)
- Principia: Relational Physics Tests for Video Models (NeurIPS 2026 · similarity 0.84)
- Learning Physics From Video: Unsupervised Physical Parameter Estimation for Continuous Dynamical Systems (CVPR 2025 · similarity 0.84)
- FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity (CVPR 2025 · similarity 0.84)
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.