Research map / 3D vision / 3D generation
PhysTwin: Physics-Informed Reconstruction and Simulation of Deformable Objects from Videos
Hanxiao Jiang, Hao-Yu Hsu, Kaifeng Zhang, Hsin-Ni Yu, Shenlong Wang, Yunzhu Li
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 hierarchical sparse-to-dense optimization (zero-order then first-order) over zero-order only (ablation) | PhysTwin · Future Prediction | CD: 0.014 -> 0.012 | Improved |
| add hierarchical sparse-to-dense optimization over zero-order only (ablation) | PhysTwin · Reconstruction & Resimulation | IoU %: 80.2 -> 84.4 | Improved |
| add zero-order sparse optimization of topology on top of first-order only (ablation) | PhysTwin · Future Prediction | CD: 0.019 -> 0.012 | Improved |
| add zero-order sparse optimization on top of first-order only (ablation) | PhysTwin · Reconstruction & Resimulation | IoU %: 82.7 -> 84.4 | Improved |
| add first-order dense stiffness optimization on top of zero-order only (ablation) | PhysTwin · Future Prediction | IoU %: 69.2 -> 72.5 | Improved |
| use PhysTwin instead of GS-Dynamics | reconstruction and resimulation · Reconstruction & Resimulation | IoU %: 72.1 -> 84.4 | Improved |
| use PhysTwin instead of Spring-Gaus | future prediction · Future Prediction | CD: 0.062 -> 0.012 | Improved |
| use PhysTwin instead of GS-Dynamics | future prediction · Future Prediction | PSNR: 22.540 -> 25.617 | Improved |
| use PhysTwin instead of GS-Dynamics | generalization to unseen interactions · Generalization to Unseen Interactions | CD: 0.029 -> 0.013 | Improved |
| use PhysTwin instead of GS-Dynamics | generalization to unseen interactions · Generalization to Unseen Interactions | PSNR: 25.053 -> 26.199 | Improved |
| train GS-Dynamics with 30 data points instead of 1 | GS-Dynamics · rope double-hand stretching and folding | performance | No clear effect |
| optimize physics then use tracking after physics-constrained optimization instead of raw CoTracker3 tracks | PhysTwin · tracking | tracking quality | Improved |
Later papers that compared with this method
2 later accepted papers report PhysTwin as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- PhysHanDI: Physics-Based Reconstruction of Hand-Deformable Object Interactions (ICML 2026 · better on 15 of 16 shared results)
- NeuSpring: Neural Spring Fields for Reconstruction and Simulation of Deformable Objects from Videos (AAAI 2026 · better on 8 of 8 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PhysGen3D: Crafting a Miniature Interactive World from a Single Image (CVPR 2025 · similarity 0.90)
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · similarity 0.89)
- Vid2Sim: Generalizable, Video-based Reconstruction of Appearance, Geometry and Physics for Mesh-free Simulation (CVPR 2025 · similarity 0.88)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.88)
- MonoPhysics: Estimating Geometry, Appearance, and Physical Parameters from Monocular Videos (NeurIPS 2026 · similarity 0.88)
- Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image (NeurIPS 2026 · similarity 0.87)
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.
