Research map / 3D vision / 3D generation
Learning 3D-Gaussian Simulators from RGB Videos
Mikel Zhobro, Andreas René Geist, Georg Martius
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 Temporal Encoding and Merging (TEM) module (ablation) | 3DGSim · Elastic dataset | PSNR (future): 18.19 -> 33.15 | Improved |
| use latent implicit particle features instead of explicit 3DGS primitives (ablation) | 3DGSim · Elastic dataset | PSNR (future): 29.69 -> 33.15 | Improved |
| train with 12-step future rollout instead of 2-step (ablation) | 3DGSim · Elastic dataset | PSNR (future): 26.43 -> 33.15 | Improved |
| use 4 past input steps instead of 2 (ablation) | 3DGSim · Elastic dataset | PSNR (future): 32.05 -> 33.15 | Improved |
| predict deltas only for the last step instead of all 4 steps (ablation) | 3DGSim · Elastic dataset | PSNR (future): 31.98 -> 33.15 | Improved |
| use latent representation with fewer cameras (3 of 6) (ablation) | 3DGSim · Elastic dataset, camera ablation | PSNR (future): 21.02 -> 31.60 | Improved |
| add static-background segmentation masks (ablation) | 3DGSim · Elastic dataset | PSNR (future): 32.66 -> 33.15 | No clear effect |
| use grid size 0.004 instead of 0.002 or 0.008 (ablation) | 3DGSim · Elastic dataset | PSNR (future): 25.39 -> 33.15 | Improved |
| add learned timestep positional embedding in temporal merging (ablation) | 3DGSim · Elastic dataset | PSNR (future): 25.07 -> 33.15 | Improved |
| merge temporal information after early spatial stages ([1,2,2..]) rather than at the start or too late (ablation) | 3DGSim · Elastic dataset | PSNR (future): 18.87 -> 33.15 | Improved |
| use 8 past frames at inference when trained with 4 (ablation) | 3DGSim 4-12 latent · Elastic dataset | PSNR (future): 33.15 -> 28.60 | Hurt |
| use 3DGSim instead of fine-tuned Cosmos video model | CosmosFT · Elastic dataset | PSNR (future): 26.97 -> 33.15 | Improved |
Closest papers by meaning
- Learning 3D Particle-based Simulators from RGB-D Videos (ICLR 2024 · similarity 0.92)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.88)
- FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity (CVPR 2025 · similarity 0.88)
- FlexGS: Train Once, Deploy Everywhere with Many-in-One Flexible 3D Gaussian Splatting (CVPR 2025 · similarity 0.88)
- SfM-Free 3D Gaussian Splatting via Hierarchical Training (CVPR 2025 · similarity 0.88)
- HiCoM: Hierarchical Coherent Motion for Dynamic Streamable Scenes with 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.88)
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.
