Research map / Video / Video generation
Composing Driving Worlds through Disentangled Control for Adversarial Scenario Generation
Yifan Zhan, Zhengqing Chen, Qingjie Wang, Zhuo He, Muyao Niu, Xiaoyang Guo, Wei Yin, Weiqiang Ren, Qian Zhang, zheng yinqiang
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 |
|---|---|---|---|
| use CompoSIA instead of MagicDrive-V2 for 6-view scene following | CompoSIA · nuScenes 6v scene following | FVD: 152.80 -> 133.66 | Improved |
| CompoSIA identity injection vs LoRA-Edit for identity control | CompoSIA · Identity control | FVD: 161.32 -> 149.15 | Improved |
| CompoSIA action conditioning vs MagicDrive-V2 | CompoSIA · Action following | TransErr (x1000): 13.66 -> 7.37 | Improved |
| CompoSIA vs MagicDrive-V2 under action editing | CompoSIA · Action editing | TransErr (x1000): 22.86 -> 12.15 | Improved |
| add local residual modulation (AdaLN) branch to action conditioning (ablation) | CompoSIA · Action following | RotErr: 2.84 -> 0.55 | Improved |
| add local residual modulation (AdaLN) branch to action conditioning (ablation) | CompoSIA · Action following | TransErr (x1000): 15.80 -> 7.37 | Improved |
| add global PRoPE attention branch to action conditioning (ablation) | CompoSIA · Action following | TransErr (x1000): 11.24 -> 7.37 | Improved |
| add global PRoPE attention branch to action conditioning (ablation) | CompoSIA · Action following | RotErr: 0.62 -> 0.55 | Improved |
| stop identity injection earlier (T_id=0.6) vs later (T_id=0.2) (ablation) | CompoSIA identity injection · Identity control (qualitative) | identity similarity vs editing freedom | Mixed |
| remove structure conditioning at sampling (ablation) | CompoSIA · Qualitative ablation | surrounding vehicle motion consistency | Improved |
| inject identity cues throughout denoising rather than first-frame reference only (ablation) | CompoSIA · Tunnel illumination scenes (qualitative) | identity drift | Improved |
| generated video instead of ground truth as planner input | Epona planner · nuScenes planning | Avg collision rate (%): 0.35 -> 0.59 | Hurt |
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.
- MagicDrive-V2 — MagicDrive-V2: High-Resolution Long Video Generation for Autonomous Driving with Adaptive Control (ICCV 2025) · better on 6 of 6 shared results
- ReCamMaster — ReCamMaster: Camera-Controlled Generative Rendering from A Single Video (ICCV 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
- HorizonForge: Driving Scene Editing with Any Trajectories and Any Vehicles (CVPR 2026 · similarity 0.88)
- Panacea: Panoramic and Controllable Video Generation for Autonomous Driving (CVPR 2024 · similarity 0.87)
- X-Scene: Large-Scale Driving Scene Generation with High Fidelity and Flexible Controllability (NeurIPS 2025 · similarity 0.87)
- Generating Multimodal Driving Scenes via Next-Scene Prediction (CVPR 2025 · similarity 0.87)
- ReSim: Reliable World Simulation for Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- Scenario Dreamer: Vectorized Latent Diffusion for Generating Driving Simulation Environments (CVPR 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.
