Research map / Video / Video generation
Rethinking Driving World Model as Synthetic Data Generator for Perception Tasks
Kai Zeng, Zhanqian Wu, Kaixin Xiong, Xiaobao Wei, Xiangyu Guo, Zhenxin Zhu, Kalok Ho, Lijun Zhou, Bohan Zeng, Ming Lu, Haiyang Sun, BING WANG and 3 more
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 |
|---|---|---|---|
| train on real plus synthetic data from earlier driving world models at equal epochs rather than pretrain then finetune | prior world-model augmentation · nuScenes detection 2x epochs | mAP: 38.4 (real) vs 38.1 (SubjectDrive) | No clear effect |
| add 420 Dream4Drive synthetic samples (<2%) to real data | perception detector · nuScenes detection 256x512, 1x epochs | mAP: 34.5 -> 36.1 | Improved |
| add Dream4Drive synthetic data at equal 2x epochs | perception detector · nuScenes detection 256x512, 2x epochs | NDS: 50.4 -> 50.6 | Improved |
| add Dream4Drive synthetic data at 512x768 | perception detector · nuScenes detection 512x768, 1x epochs | mAP: 36.1 -> 40.7 | Improved |
| use generative 3D-aware rendering instead of naive projection of 3D assets (ablation) | Dream4Drive · nuScenes detection 512x768, 3x epochs | mAP: 43.1 -> 44.5 | Improved |
| use generative rendering instead of naive insertion (ablation) | Dream4Drive · nuScenes detection 512x768, 2x epochs | mAOE: 37.5 -> 39.4 | Hurt |
| insert assets on left side rather than right side (ablation) | Dream4Drive · nuScenes detection 512x768, 1x epochs | mAOE: 51.4 -> 45.7 | Improved |
| insert assets at far distance instead of close (ablation) | Dream4Drive · nuScenes detection 512x768, 1x epochs | mAP: 39.7 -> 40.5 | Improved |
| use multi-view generated DriveObj3D assets instead of Trellis assets (ablation) | Dream4Drive · nuScenes detection 512x768, 1x epochs | mAP: 39.8 -> 40.7 | Improved |
| use multi-view assets instead of single-view Hunyuan3D assets (ablation) | Dream4Drive · nuScenes detection 512x768, 1x epochs | mAP: 40.2 -> 40.7 | Improved |
| increase OOD scenes from 7 to 35 (ablation) | Dream4Drive · nuScenes detection 512x768, 1x epochs | mAP: 40.7 -> 39.7 | Hurt |
| add style-transferred (rain, night) synthetic data (ablation) | Dream4Drive · nuScenes detection 512x768, 2x epochs | mAP: 43.6 -> 44.2 | 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.
- Panacea — Panacea: Panoramic and Controllable Video Generation for Autonomous Driving (CVPR 2024) · better on 2 of 2 shared results
- DriveDreamer — DriveDreamer: Towards Real-world-driven World Models for Autonomous Driving (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 0.91)
- DriveScape: High-Resolution Driving Video Generation by Multi-View Feature Fusion (CVPR 2025 · similarity 0.89)
- SubjectDrive: Scaling Generative Data in Autonomous Driving via Subject Control (AAAI 2025 · similarity 0.89)
- DriveDreamer-2: LLM-Enhanced World Models for Diverse Driving Video Generation (AAAI 2025 · similarity 0.88)
- DrivingSphere: Building a High-fidelity 4D World for Closed-loop Simulation (CVPR 2025 · similarity 0.88)
- Glad: A Streaming Scene Generator for Autonomous Driving (ICLR 2025 · 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.
