Research map / 3D vision / 3D generation
Future Dynamic 3D Reconstruction: A 3D World Model with Disentangled Ego-Motion
Nils Morbitzer, Jonathan Evers, Artem Savkin, Thomas Stauner, Nassir Navab, Federico Tombari, Stefano Gasperini
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 information sharing between pose and spatial masked transformers via cross-attention (ablation) | FR3D (A3 baseline) · Waymo 224x224, t+2.0s pose | ATE: 0.895 -> 0.389 | Improved |
| add information sharing between pose and spatial transformers (ablation) | FR3D (A3 baseline) · Waymo 224x224, t+2.0s depth | AbsRel: 0.183 -> 0.173 | Improved |
| add autoregressive training (ablation) | FR3D (A3 baseline) · Waymo 224x224, t+2.0s depth | AbsRel: 0.183 -> 0.179 | Improved |
| add autoregressive training (ablation) | FR3D (A3 baseline) · Waymo 224x224, t+2.0s pose | ATE / RPE_R: ATE 0.895 -> 0.670, RPE_R 0.724 -> 0.988 | Mixed |
| combine autoregressive training and information sharing (ablation) | FR3D · Waymo 224x224, t+2.0s depth | AbsRel: 0.173 -> 0.158 (A5 -> A6) | Improved |
| combine autoregressive training and information sharing (ablation) | FR3D · Waymo 224x224, t+2.0s pose | ATE: 0.389 -> 0.442 (A5 -> A6) | Mixed |
| add pose masked transformer to CUT3R-Foresight (ablation) | CUT3R-Foresight · Waymo 224x224 depth | AbsRel: 0.183 -> 0.183 at t+2.0s | No clear effect |
| fine-tune at higher resolution 512x336 instead of 224x224 (ablation) | FR3D · Waymo depth, t+2.0s | AbsRel: 0.158 -> 0.152 | Improved |
| use CUT3R latent tokens with disentangled pose forecasting instead of DINOv2 features (DINO-Foresight) | future depth forecasting · nuScenes zero-shot, t+2.5s | AbsRel: 0.283 -> 0.229 | Improved |
| disentangle ego-motion and forecast with FR3D instead of CUT3R-Foresight on static regions | future depth forecasting · Waymo static regions, t+2.0s | AbsRel: 0.170 -> 0.145 | Improved |
| disentangle ego-motion and forecast with FR3D instead of CUT3R-Foresight on dynamic regions | future depth forecasting · Waymo dynamic regions, t+2.0s | AbsRel: 0.353 -> 0.337 | Improved |
| forecast with FR3D rather than a 2D video world model (Vista) | future prediction · 2s forecast, 4 context frames | latency (s): 24.57 -> 0.85 | Improved |
Closest papers by meaning
- World from Motion: Generative Dynamic Gaussian Reconstruction from Monocular Video (NeurIPS 2026 · similarity 0.89)
- Learning 3D Persistent Embodied World Models (NeurIPS 2025 · similarity 0.89)
- World Motion Models: Flexible Sequence Modeling of SE(3) Trajectories (NeurIPS 2026 · similarity 0.88)
- DynamicVerse: A Physically-Aware Multimodal Framework for 4D World Modeling (NeurIPS 2025 · similarity 0.87)
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 2025 · similarity 0.87)
- Scalable Benchmarking and Robust Learning for Noise-Free Ego-Motion and 3D Reconstruction from Noisy Video (ICLR 2025 · 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.
