Research map / Video / Video generation
GenieDrive: Towards Physics-Aware Driving World Model with 4D Occupancy Guided Video Generation
Zhenya Yang, Zhe Liu, Yuxiang Lu, Liping Hou, Chenxuan Miao, Siyi Peng, Bailan Feng, Xiang Bai, Hengshuang Zhao
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 Mutual Control Attention (ablation) | GenieDrive occupancy world model · nuScenes occupancy forecasting | Avg. mIoU (%): 38.96 -> 42.59 | Improved |
| add end-to-end training of VAE and predictor (ablation) | GenieDrive occupancy world model · nuScenes occupancy forecasting | Avg. mIoU (%): 39.79 -> 42.59 | Improved |
| add continuous representation (vs vector quantization) (ablation) | GenieDrive occupancy world model · nuScenes occupancy forecasting | Avg. mIoU (%): 39.09 -> 42.59 | Improved |
| add continuous representation and E2E training together (ablation) | GenieDrive occupancy world model · nuScenes occupancy forecasting | Avg. mIoU (%): 35.24 -> 42.59 | Improved |
| apply end-to-end training to DOME (ablation) | DOME · nuScenes occupancy forecasting | Avg. mIoU (%): 27.10 -> 0.43 | Hurt |
| apply end-to-end training to I2-World (ablation) | I2-World · nuScenes occupancy forecasting | Avg. mIoU (%): 39.73 -> 10.09 | Hurt |
| add Normalized Multi-View Attention (ablation) | GenieDrive video generator · nuScenes multi-view video (37 frames) | FVD: 120.16 -> 98.06 | Improved |
| add cross normalization to multi-view attention (ablation) | GenieDrive video generator · nuScenes multi-view video (37 frames) | FVD: 212.67 -> 98.06 | Improved |
| add cross normalization to multi-view attention (ablation) | GenieDrive video generator · nuScenes multi-view video (37 frames) | mAP (%): 10.04 -> 19.84 | Improved |
| longer E2E training epochs (0 -> 4 epochs) (ablation) | GenieDrive occupancy world model · nuScenes forecasting vs reconstruction | reconstruction mIoU / forecasting mIoU: 86.15 -> 73.89 reconstruction mIoU; 39.79 -> 41.64 forecasting mIoU | Mixed |
| tri-plane GenieDrive vs I2-World | GenieDrive occupancy world model · nuScenes occupancy forecasting | Avg. mIoU (%): 39.73 -> 42.59 | Improved |
| tri-plane VAE vs I2-World on reconstruction IoU | GenieDrive occupancy world model · nuScenes occupancy forecasting | Reconstruction IoU (%): 77.25 -> 75.53 | Hurt |
Closest papers by meaning
- DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation (CVPR 2025 · similarity 0.88)
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.88)
- 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)
- Driving into the Future: Multiview Visual Forecasting and Planning with World Model for Autonomous Driving (CVPR 2024 · similarity 0.87)
- DriveScape: High-Resolution Driving Video Generation by Multi-View Feature Fusion (CVPR 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.
