MetaOcc

<p align="center"> <h3 align="center"><strong>MetaOcc: Spatio-Temporal Fusion of Surround-View 4D Radar and Camera for 3D Occupancy Prediction with Dual Training Strategies</strong></h3> </p> <p align="center"> Long Yang^*&nbsp;&nbsp;&nbsp; <a href="https://zhenglianqing.github.io/">Lianqing Zheng</a>^*&nbsp;&nbsp;&nbsp; Wenjin Ai</sup>&nbsp;&nbsp;&nbsp; Minghao Liu</sup>&nbsp;&nbsp;&nbsp; Sen Li</sup>&nbsp;&nbsp;&nbsp; Qunshu Lin</sup>&nbsp;&nbsp;&nbsp; Shengyu Yan</sup>&nbsp;&nbsp;&nbsp; Jie Bai</sup>&nbsp;&nbsp;&nbsp; Zhixiong Ma</sup>&nbsp;&nbsp;&nbsp; Tao Huang</sup>&nbsp;&nbsp;&nbsp; Xichan Zhu</sup>&nbsp;&nbsp;&nbsp; </p> <p align="center"> <a href="https://arxiv.org/pdf/2501.15384" target='_blank'> <img src="https://img.shields.io/badge/Paper-%F0%9F%93%83-purple"> </a> <a href="https://github.com/TJRadarLab/OmniHD-Scenes" target='_blank'> <img src="https://img.shields.io/badge/OmniHD--Scenes-📦-violet"> </a> <a href="https://github.com/weiyithu/SurroundOcc/blob/main/docs/data.md" target='_blank'> <img src="https://img.shields.io/badge/SurroundOcc--nuScenes-📦-blu"> </a> </p>

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.

<div style="text-align: center;"> <img src="assets\framework.png" alt="Dialogue_Teaser" width=100% > </div>

Results

• 3D semantic occupancy prediction results on nuScenes.

<div style="text-align: center;"> <img src="assets\SurroundOcc_nuScenes_Result.png" alt="Dialogue_Teaser" width=100% > </div>

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

<div style="text-align: center;"> <img src="assets\OmniHD_Scenes_Result.png" alt="Dialogue_Teaser" width=100% > </div>

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!