Research map / 3D vision / 3D generation
Geometry-Aware Spatio-Temporal Context Modeling for 4D Occupancy Forecasting
Sitao Chen, Zhuangwei Zhuang, Hui Luo, Qingyao Wu, Mingkui Tan
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 explicit geometric transformation (EGT) (ablation) | per-frame baseline · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 18.31 -> 31.78 | Improved |
| add implicit feature modulation (IFM) (ablation) | baseline+EGT · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 31.78 -> 32.55 | Improved |
| add feature refinement (FR) (ablation) | baseline+EGT+IFM · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 32.55 -> 40.47 | Improved |
| add global spatial context aggregation (GSCA) (ablation) | PEIG · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 40.47 -> 46.08 | Improved |
| add temporal dynamics extraction (TDE) (ablation) | PEIG+GSCA · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 46.08 -> 47.40 | Improved |
| use 4 history frames instead of 1 (predicted ego) (ablation) | GAST · Occ3D-nuScenes val (3D-Occ, predicted ego) | mIoU (%): 17.01 -> 27.38 | Improved |
| use 4 history frames instead of 1 (GT ego) (ablation) | GAST · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 34.88 -> 47.40 | Improved |
| stack more DCA layers in feature refinement (1 to 3) (ablation) | GAST · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 47.40 -> 47.71 | No clear effect |
| increase ConvGRU layers from 1 to 4 (ablation) | GAST · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 47.03 -> 47.45 | Improved |
| use multi-scale BEV (50, 100, 200) instead of single 50x50 (ablation) | GSCA · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 46.90 -> 47.40 | Improved |
| use multi-scale BEV instead of single 100x100 (ablation) | GSCA · Occ3D-nuScenes val (3D-Occ, GT ego) | mIoU (%): 47.18 -> 47.40 | Improved |
| add historical occupancy reconstruction loss (ablation) | GAST · Occ3D-nuScenes val | mIoU (%) | 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.
- I2-World — I2-World: Intra-Inter Tokenization for Efficient Dynamic 4D Scene Forecasting (ICCV 2025) · better on 7 of 7 shared results
- COME — COME: Adding Scene-Centric Forecasting Control to Occupancy World Model (NeurIPS 2025) · better on 4 of 4 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
- OccProphet: Pushing the Efficiency Frontier of Camera-Only 4D Occupancy Forecasting with an Observer-Forecaster-Refiner Framework (ICLR 2025 · similarity 0.91)
- Spatiotemporal Decoupling for Efficient Vision-Based Occupancy Forecasting (CVPR 2025 · similarity 0.91)
- Cam4DOcc: Benchmark for Camera-Only 4D Occupancy Forecasting in Autonomous Driving Applications (CVPR 2024 · similarity 0.91)
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.90)
- UnO: Unsupervised Occupancy Fields for Perception and Forecasting (CVPR 2024 · similarity 0.89)
- COME: Adding Scene-Centric Forecasting Control to Occupancy World Model (NeurIPS 2025 · similarity 0.89)
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.
