Research map / 3D vision / 3D generation
SceneDirector: Bridging Explicit Geometry and Generative Priors for Unified Driving Scene Editing
Yiyuan Liang, Zhiying Yan, Tao Zhang, Shangke Liu, Kai Lin, Xu Zou, Nong Sang, Changxin Gao
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 Mask-Gated Reference Attention (reference injection) (ablation) | SceneDirector · multi-view 3m deviation | FVD: 495.0 -> 469.2 | Improved |
| add Mask-Gated Reference Attention (reference injection) (ablation) | SceneDirector · multi-view 3m deviation | FID: 53.02 -> 49.10 | Improved |
| add Mask-Gated Reference Attention (reference injection) (ablation) | SceneDirector · multi-view 2m fixed offset | FVD: 512.5 -> 474.1 | Improved |
| add uncertainty-aware gating to cross-attention (MGRA vs standard attention) (ablation) | SceneDirector · multi-view 3m deviation | FVD: 481.8 -> 469.2 | Improved |
| add uncertainty-aware gating to cross-attention (MGRA vs standard attention) (ablation) | SceneDirector · multi-view 2m fixed offset | FVD: 493.4 -> 474.1 | Improved |
| replace depth completion backbone with LRRU (ablation) | SceneDirector · multi-view 2m fixed offset | FID: 51.34 -> 51.87 | No clear effect |
| replace depth completion backbone with LRRU (ablation) | SceneDirector · multi-view 3m deviation | FID: 49.10 -> 50.04 | Hurt |
| add object plus ego-trajectory editing jointly vs object-only | SceneDirector · multi-view object editing | ATE: 0.90 -> 0.95 | Hurt |
| unified single-pass editing vs sequential single-edit baselines | VACE · Multi-Edit | FVD: 729.42 -> 516.83 | Improved |
| use SceneDirector instead of VACE-14B for object insertion | VACE · Single-Edit | ATE: 1.02 -> 0.78 | Improved |
| use SceneDirector instead of VACE-14B for object insertion | VACE · Single-Edit | FVD: 451.31 -> 464.56 | Hurt |
| train StreetGaussian on SceneDirector-generated videos | StreetGaussian · front-view 2m gradual deviation | FVD: 434.9 -> 400.1 | 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.
- DriveEditor — DriveEditor: A Unified 3D Information-Guided Framework for Controllable Object Editing in Driving Scenes (AAAI 2025) · better on 4 of 4 shared results
- FreeVS — FreeVS: Generative View Synthesis on Free Driving Trajectory (ICLR 2025) · better on 2 of 2 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
- DriveEditor: A Unified 3D Information-Guided Framework for Controllable Object Editing in Driving Scenes (AAAI 2025 · similarity 0.90)
- SceneCrafter: Controllable Multi-View Driving Scene Editing (CVPR 2025 · similarity 0.90)
- X-Scene: Large-Scale Driving Scene Generation with High Fidelity and Flexible Controllability (NeurIPS 2025 · similarity 0.88)
- UniScene: Unified Occupancy-centric Driving Scene Generation (CVPR 2025 · similarity 0.87)
- Generating Multimodal Driving Scenes via Next-Scene Prediction (CVPR 2025 · similarity 0.86)
- Genesis: Multimodal Driving Scene Generation with Spatio-Temporal and Cross-Modal Consistency (NeurIPS 2025 · similarity 0.86)
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.
