Research map / 3D vision / Gaussian splatting
Learning Physics-Grounded 4D Dynamics with Neural Gaussian Force Fields
Shiqian Li, Ruihong Shen, Junfeng Ni, Chang Pan, Chi Zhang, Yixin Zhu
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 soft-body local stress field (deformation modeling) (ablation) | NGFF · GSCollision spatial | RMSE: 0.110 -> 0.082 | Improved |
| add soft-body local stress field (deformation modeling) (ablation) | NGFF · GSCollision temporal | RMSE: 0.144 -> 0.107 | Improved |
| add soft-body local stress field (deformation modeling) (ablation) | NGFF · GSCollision compositional | RMSE: 0.131 -> 0.104 | Improved |
| add soft-body local stress field (deformation modeling) (ablation) | NGFF · GSCollision compositional | correlation R: 0.515 -> 0.571 | Improved |
| add soft-body local stress field (ablation) | NGFF · GSCollision | inference time (s): 0.303 -> 0.363 | Hurt |
| learned force fields instead of Pointformer | NGFF vs Pointformer · GSCollision spatial | RMSE: 0.096 -> 0.082 | Improved |
| learned force fields instead of Pointformer | NGFF vs Pointformer · GSCollision | inference time (s): 0.183 -> 0.363 | Hurt |
| replace MPM with VLM-estimated params by learned force field | NGFF vs VLM-MPM · GSCollision | inference time (s): 39.29 -> 0.363 | Improved |
| NGFF-V rendering vs Veo3 on photorealism | NGFF-V · GSCollision video generation (All, VLM eval) | photorealism: 0.41 -> 0.35 | Hurt |
| NGFF-V vs Cosmos tuned physical realism | NGFF-V · GSCollision video generation (All) | physical realism: VLM 0.24 -> 0.30; human 0.59 -> 0.55 | Mixed |
| use NGFF-V vs Cosmos for physical realism | NGFF-V · GSCollision video generation (All, VLM eval) | physical realism: 0.20 -> 0.30 | Improved |
Closest papers by meaning
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.89)
- Force Prompting: Video Generation Models Can Learn And Generalize Physics-based Control Signals (NeurIPS 2025 · similarity 0.88)
- FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity (CVPR 2025 · similarity 0.88)
- NeuMA: Neural Material Adaptor for Visual Grounding of Intrinsic Dynamics (NeurIPS 2024 · similarity 0.88)
- OmniPhysGS: 3D Constitutive Gaussians for General Physics-Based Dynamics Generation (ICLR 2025 · similarity 0.87)
- 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering (CVPR 2024 · 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.
