Occ3D-nuScenes-SatExt Dataset Introduction

Introduction

The Occ3D-nuScenes-SatExt dataset (from SA-Occ) is an extension of the Occ3D-nuScenes dataset, integrating satellite imagery with real-time ground-level sensor data to enhance 3D occupancy prediction tasks. This dataset is the first to systematically incorporate satellite data into real-time applications using GPS and IMU for alignment, making significant contributions to the field of autonomous driving.

Relationship to Occ3D-nuScenes

Building upon the core structure and annotations of the Occ3D-nuScenes dataset, Occ3D-nuScenes-SatExt introduces satellite imagery as a new modality while maintaining compatibility with existing research frameworks and evaluation metrics.

Key Contributions

Satellite and Real-Time Data Fusion: This dataset pioneers the integration of high-resolution satellite imagery with real-time LiDAR, camera, and radar data from nuScenes. Aligning satellite images with ground-level sensor data using GPS and IMU inputs enables models to leverage both aerial context and immediate environmental details for more accurate 3D scene understanding.
Automatic Geospatial Alignment: Through precise GPS and IMU integration, Occ3D-nuScenes-SatExt achieves geospatial alignment automatically between satellite imagery and ground sensor data, essential for applications requiring precise localization.
Enhanced Large Time Span Modeling: Our satellite data, sourced from Google Earth in 2024, combined with the 2019-collected nuScenes data, provides a realistic scenario for modeling temporal changes over a large time span. This is crucial for predicting occupancy in dynamic urban environments.
Richer Semantic Context: Satellite images provide a broader field of view, capturing elements such as road layouts and building structures that might be obscured at ground level, helping models make more informed predictions about occluded areas.

Applications

The Occ3D-nuScenes-SatExt dataset is a valuable resource for advancing research in autonomous driving, particularly in 3D scene understanding, long-term prediction, and multi-sensor fusion. It supports the development of more robust and accurate perception systems for autonomous vehicles by providing comprehensive and context-rich data.

Citation

If you use this dataset or the related paper in your research, please cite it using the following format:

@article{chen2025saocc,
  title={SA-Occ: Satellite-Assisted 3D Occupancy Prediction in Real World},
  author={Chen Chen and Zhirui Wang and Taowei Sheng and Yi Jiang and Yundu Li and Peirui Cheng and Luning Zhang and Kaiqiang Chen and Yanfeng Hu and Xue Yang and Xian Sun},
  journal={arXiv preprint arXiv:2503.16399},
  year={2025}
}

Datasets:

chenchen235
/

Occ3D_nuScenes_SatExt