Research map / Video / Video segmentation and tracking
Pixel2Phys: Distilling Governing Laws from Visual Dynamics
Ruikun Li, Jun Yao, Yingfan Hua, Shixiang Tang, Biqing Qi, Bin Liu, Wanli Ouyang, Yan Lu
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 |
|---|---|---|---|
| replace implicit latent dynamics (AE-SINDy, Latent-ODE) with explicit object-level parsing in Pixel2Phys | Pixel2Phys · Object motion (5 systems) | R2@1000: 0.0046 -> 0.9913 (Linear, AE-SINDy vs Pixel2Phys) | Improved |
| use Pixel2Phys instead of Coord-Equ for equation discovery | object motion videos · Cubic | false positives: 3.400 -> 0.3900 | Improved |
| use Pixel2Phys instead of Coord-Equ | Circular motion · Circular | R2@1000: 0.9903 -> 1.0000 | Improved |
| use Pixel2Phys instead of black-box neural operators on PDE videos | Lambda-Omega · Lambda-Omega | RMSE: 0.42 -> 0.03 (UNO vs Pixel2Phys) | Improved |
| use Pixel2Phys instead of SGA-PDE | Brusselator · Brusselator | RMSE: 0.14 -> 0.12 | Improved |
| use Pixel2Phys instead of PDE-Find | Swift-Hohenberg · Swift-Hohenberg | RMSE: 0.19 -> 0.18 | No clear effect |
| use Pixel2Phys instead of FNO/PDE-Find on VPS | Swift-Hohenberg · Swift-Hohenberg | VPS@0.5: 200.00 -> 200.00 (PDE-Find vs Pixel2Phys) | No clear effect |
| replace Plan Agent with a static serial workflow (no equation feedback loop) (ablation) | Pixel2Phys · variable space quality (Appendix F.2) | parsimony of discovered laws | Hurt |
| substitute GPT-4o backbone with smaller-scale LLMs (ablation) | Pixel2Phys · Appendix F.3 | accuracy | No clear effect |
| use Pixel2Phys instead of Wan2.2 video generation on real vortex videos | representation-level videos · Karman vortex and BZ videos | RMSE / vorticity error | Improved |
Closest papers by meaning
- Mimicking the Physicist's Eye: A VLM-centric Approach for Physics Formula Discovery (NeurIPS 2026 · similarity 0.87)
- Apple-$\pi$: Benchmarking Thinking with Video Towards Law-Grounded Physical Intelligence (NeurIPS 2026 · similarity 0.87)
- Beyond Static Vision: Scene Dynamic Field Unlocks Intuitive Physics Understanding in Multi-modal Large Language Models (ICLR 2026 · similarity 0.87)
- WISA: World simulator assistant for physics-aware text-to-video generation (NeurIPS 2025 · similarity 0.86)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.86)
- PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World 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.
