Research map / Vision-language and multimodal / Human and embodied modeling
Learning to See Physics via Visual De-animation
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 joint end-to-end fine-tuning (init -> full) (ablation) | VDA · Billiards I | position error 20th frame (px): 26.10 -> 25.55 | Helped |
| add joint end-to-end fine-tuning (init -> full) (ablation) | VDA · Billiards II | position error 20th frame (px): 34.65 -> 32.97 | Helped |
| add joint end-to-end fine-tuning (init -> full) (ablation) | VDA · Billiards III | position error 20th frame (px): 46.40 -> 43.42 | Helped |
| add joint end-to-end fine-tuning (init -> full) (ablation) | VDA · Billiards II | reconstruction pixel MSE: 0.058 -> 0.055 | Helped |
| add joint end-to-end fine-tuning (init -> full) (ablation) | VDA · Billiards II | velocity error 1st frame (px): 1.65 -> 1.64 | No clear effect |
| use VDA instead of baseline | Baseline · Billiards I | position error 20th frame (px): 119.97 -> 25.55 | Helped |
| use VDA instead of baseline | Baseline · Billiards III | position error 20th frame (px): 202.56 -> 43.42 | Helped |
| use VDA instead of baseline | Baseline · Billiards II | velocity error 20th frame (px): 12.02 -> 3.05 | Helped |
| add joint REINFORCE fine-tuning on stability labels without stability constraint (ablation) | VDA init · Blocks stability | mean accuracy (%): 72 -> 74 | Helped |
| add physical stability constraint during fine-tuning (ablation) | VDA joint · Blocks stability | mean accuracy (%): 74 -> 75 | Helped |
| use VDA full instead of PhysNet | PhysNet · Blocks stability | mean accuracy (%): 65 -> 75 | Helped |
| use VDA full instead of PhysNet on 4-block towers | PhysNet · Blocks stability 4 blocks | accuracy (%): 70 -> 72 | Helped |
Closest papers by meaning
- Physics-as-Inverse-Graphics: Unsupervised Physical Parameter Estimation from Video (ICLR 2020 · similarity 0.89)
- Dynamic Visual Reasoning by Learning Differentiable Physics Models from Video and Language (NeurIPS 2021 · similarity 0.88)
- Latent Intuitive Physics: Learning to Transfer Hidden Physics from A 3D Video (ICLR 2024 · similarity 0.88)
- PhysGen3D: Crafting a Miniature Interactive World from a Single Image (CVPR 2025 · similarity 0.87)
- PhysicsGen: Can Generative Models Learn from Images to Predict Complex Physical Relations? (CVPR 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)
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.