Research map / 3D vision / 3D generation
Fast-Slow Evolutionary Occupancy Prediction via Controlled Dynamics
Runhe Yang, Zhiyuan Zhou, Yuxiang Yan, Xuetong Yang, Yikai Pan, Jiacheng Tang, Zhuolin He, Jianhua Han, Hang Xu, Jian Pu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a fast-slow evolutionary framework for 3D semantic occupancy prediction from driving sensors that aims to combine high accuracy with low latency.
Problem
3D semantic occupancy prediction for autonomous driving faces a trade-off: performance-oriented models are accurate but slow, while deployment-friendly models are fast but less accurate.
Approach
Uses a fast-slow evolutionary scheme with controlled dynamics to get accuracy and low latency together; details are not available from the abstract alone.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace control increment with elapsed time (ablation) | EvoOcc · Ablation streaming (ALOcc-2D-mini / ALOcc-3D) | mIoU / RayIoU: 40.3 -> 39.9 mIoU; 41.5 -> 41.1 RayIoU | Hurt |
| add ray-termination loss (ablation) | EvoOcc · Ablation streaming (ALOcc-2D-mini / ALOcc-3D) | RayIoU: 40.4 -> 41.5 | Improved |
| add ray-termination loss (ablation) | EvoOcc · Ablation streaming (ALOcc-2D-mini / ALOcc-3D) | mIoU: 40.6 -> 40.3 | No clear effect |
| add ray-termination loss (ablation) | EvoOcc · Surface-geometry diagnostics | extra occupancy (%): 4.30 -> 3.01 | Improved |
| use scene state evolution instead of convolution learned fusion (ablation) | EvoOcc · Ablation streaming (ALOcc-2D-mini / ALOcc-3D) | mIoU: 39.8 -> 40.3 | Improved |
| use scene state evolution instead of attention learned fusion (ablation) | EvoOcc · Ablation streaming (ALOcc-2D-mini / ALOcc-3D) | mIoU: 38.4 -> 40.3 | Improved |
| add EvoOcc evolution to fast system | ALOcc-2D-mini / ALOcc-3D · Occ3D-nuScenes val offline | mIoU: 33.4 -> 39.3 | Improved |
| ego-motion warping only of delayed slow (no scene evolution) (ablation) | delayed slow system · Occ3D-nuScenes val offline | mIoU_D: 31.3 -> 20.0 (vs fast, ALOcc pair) | Hurt |
| add EvoOcc in streaming setting | fast systems · Occ3D-nuScenes val streaming | mIoU: +4.3 to +6.9 | Improved |
| lengthen slow system invocation interval 0.5 s to 5.0 s | EvoOcc · ALOcc pair streaming | mIoU: 40.3 -> 38.9; 23.6 -> 11.0 TFLOPs/scene | Mixed |
| lengthen slow invocation interval to 20.0 s | EvoOcc · ALOcc pair streaming | mIoU: 40.3 -> 35.6 | Mixed |
| vary fast arrival rate (fixed, alternating, random) | EvoOcc · irregular arrival | mIoU_D / RayIoU: mIoU_D varies at most 0.2; RayIoU at most 0.1 | No clear effect |
Closest papers by meaning
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.89)
- COME: Adding Scene-Centric Forecasting Control to Occupancy World Model (NeurIPS 2025 · similarity 0.89)
- OccProphet: Pushing the Efficiency Frontier of Camera-Only 4D Occupancy Forecasting with an Observer-Forecaster-Refiner Framework (ICLR 2025 · similarity 0.88)
- Achieving Speed-Accuracy Balance in Vision-based 3D Occupancy Prediction via Geometric-Semantic Disentanglement (AAAI 2025 · similarity 0.88)
- GaussianWorld: Gaussian World Model for Streaming 3D Occupancy Prediction (CVPR 2025 · similarity 0.88)
- Cam4DOcc: Benchmark for Camera-Only 4D Occupancy Forecasting in Autonomous Driving Applications (CVPR 2024 · 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.
