Research map / 3D vision / 3D generation
OccWorld: Learning a 3D Occupancy World Model for Autonomous Driving
Wenzhao Zheng, Weiliang Chen, Yuanhui Huang, Borui Zhang, Yueqi Duan, Jiwen Lu
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 spatial aggregation before temporal attention (ablation) | OccWorld-O · Occ3D-nuScenes forecasting | Avg. mIoU (%): 10.07 -> 17.14 | Improved |
| use temporal attention instead of simple convolution (ablation) | OccWorld-O · Occ3D-nuScenes forecasting | Avg. mIoU (%): 8.98 -> 17.14 | Improved |
| add ego token (joint modeling of scene and ego) (ablation) | OccWorld-O · Occ3D-nuScenes forecasting | Avg. mIoU (%): 15.13 -> 17.14 | Improved |
| use temporal attention for the ego token instead of MLP (ablation) | OccWorld-O · nuScenes planning | Avg. L2 (m): 5.89 -> 1.17 | Improved |
| use codebook size 1024 instead of 512 (ablation) | scene tokenizer · Occ3D-nuScenes forecasting | Avg. mIoU (%): 17.14 -> 16.30 | Hurt |
| increase latent spatial resolution of tokenizer (ablation) | scene tokenizer · Occ3D-nuScenes forecasting | Avg. forecasting mIoU (%): reconstruction mIoU 66.38 -> 78.12; forecasting 17.14 -> 12.38 | Mixed |
| reduce tokenizer channel dimension to 64 (ablation) | scene tokenizer · Occ3D-nuScenes forecasting | Avg. mIoU (%): 17.14 -> 14.67 | Hurt |
| use ground-truth 3D occupancy instead of predicted camera occupancy as input | OccWorld · Occ3D-nuScenes forecasting | Avg. mIoU (%): 8.62 -> 17.14 | Improved |
| use world-model paradigm with GT occupancy instead of OccNet pipeline | planning · nuScenes planning | Avg. L2 (m): 2.25 -> 1.17 | Improved |
| OccWorld-O versus UniAD at long horizon planning | OccWorld-O · nuScenes planning | L2 at 3s (m): 1.99 vs 1.65 (UniAD) | Hurt |
Closest papers by meaning
- Semi-Supervised Vision-Centric 3D Occupancy World Model for Autonomous Driving (ICLR 2025 · similarity 0.93)
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.92)
- GaussianWorld: Gaussian World Model for Streaming 3D Occupancy Prediction (CVPR 2025 · similarity 0.92)
- COME: Adding Scene-Centric Forecasting Control to Occupancy World Model (NeurIPS 2025 · similarity 0.92)
- Achieving Speed-Accuracy Balance in Vision-based 3D Occupancy Prediction via Geometric-Semantic Disentanglement (AAAI 2025 · similarity 0.91)
- UnO: Unsupervised Occupancy Fields for Perception and Forecasting (CVPR 2024 · similarity 0.91)
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.
