Research map / 3D vision / 3D generation
UFO: Unifying Feed-Forward and Optimization-based Methods for Large Driving Scene Modeling
Kaiyuan Tan, Yingying Shen, Ziyue Zhu, Mingfei Tu, Haohui Zhu, Haiyang Sun, Bing Wang, Guang Chen, Hangjun Ye
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 recurrent scene tokens with refinement over independent clip prediction (ablation) | UFO · WOD 8s | PSNR: 21.25 -> 27.39 | Improved |
| add 3D-aware memory over implicit memory (ablation) | UFO · WOD 8s | PSNR: 26.67 -> 27.08 | Improved |
| add refinement of earlier tokens to 3D-aware memory (ablation) | UFO · WOD 8s | PSNR: 27.08 -> 27.39 | Improved |
| add historical scene tokens (ablation) | UFO · WOD 8s | PSNR: 26.31 -> 27.39 | Improved |
| add progressive training on longer sequences (ablation) | UFO · WOD 8s | PSNR: 20.67 -> 27.39 | Improved |
| add lifespan prediction for Gaussians (ablation) | UFO · WOD 8s | PSNR: 26.47 -> 27.39 | Improved |
| add bounding box guidance for dynamic objects (ablation) | UFO · WOD 8s | PSNR: 26.88 -> 27.39 | Improved |
| use UFO instead of STORM as sequence grows to 16s (zero-shot) | UFO · WOD 16s zero-shot | PSNR: 22.02 -> 27.04 | Improved |
| use UFO instead of STORM on short 2s sequences | UFO · WOD 2s | Depth RMSE: 5.48 -> 5.45 | No clear effect |
| use bounding-box and lifespan dynamic modeling instead of STORM constant velocity (10 frames) | UFO · WOD dynamic objects 2s | PSNR: 22.92 -> 25.08 | Improved |
| visibility-based filtering instead of global attention | UFO · WOD 16s inference | Inference time: 1.47s -> 0.44s | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- GS-LRM — GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (ECCV 2024) · better on 3 of 3 shared results
- Street Gaussians — Street Gaussians: Modeling Dynamic Urban Scenes with Gaussian Splatting (ECCV 2024) · better on 1 of 1 shared result
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
- DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation (CVPR 2025 · similarity 0.87)
- DrivingSphere: Building a High-fidelity 4D World for Closed-loop Simulation (CVPR 2025 · similarity 0.86)
- DualAD: Disentangling the Dynamic and Static World for End-to-End Driving (CVPR 2024 · similarity 0.86)
- VR-Drive: Viewpoint-Robust End-to-End Driving with Feed-Forward 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.86)
- X-Scene: Large-Scale Driving Scene Generation with High Fidelity and Flexible Controllability (NeurIPS 2025 · similarity 0.86)
- Future-Aware End-to-End Driving: Bidirectional Modeling of Trajectory Planning and Scene Evolution (NeurIPS 2025 · similarity 0.85)
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.
