MetaOcc: Spatio-Temporal Fusion of Surround-View 4D Radar and Camera for 3D Occupancy Prediction with Dual Training Strategies

Long Yang^*    Lianqing Zheng^*    Wenjin Ai    Minghao Liu    Sen Li    Qunshu Lin    Shengyu Yan    Jie Bai    Zhixiong Ma    Tao Huang    Xichan Zhu   

Overview

This repository is an official implementation of MetaOcc, the first framework to fuse surround-view 4D radar and camera data for robust and annotation-efficient 3D occupancy prediction in autonomous driving.

Results

• 3D semantic occupancy prediction results on nuScenes.

• 3D semantic occupancy prediction results on OmniHD-Scenes.

Updates

[2025/8/12] 🎉 OmniHD-Scenes dataset v1.0 (~1.3TB) is openly for research purposes. Apply for download here and explore it via the official repository.

[2025/1/31] 🚀 Our paper is available on arXiv. Code will be released soon!