AutoMine Dataset


AutoMine is the first autonomous driving dataset for perception and localization in mine scenarios which is unstructured special scenarios. AutoMine can provide data assistance for research in autonomous driving, computer vision, deep learning, and so on. It is freely open to non-commercial use.


At present, AutoMine supports two major autonomous driving tasks. More tasks will be taken into account in the future.


w to maximum learning potential.

We offer abundant vehicle location information based on GPS and IMU. We recommend users utilize the lidar-based, lidar-inertia SLAM (Simultaneous Localization and Mapping), pure vision, vision-inertia SLAM, and fusion SLAM methods on our dataset.


There are 36k frames of data in the whole perception dataset (12k frames per platform on average). We annotated data provides roll and pitch angles, especially pitch angles. We hope that researchers can explore all 9 DoF (degrees of freedom), with special attention to roll and pitch, which is essential for safe driving in strip mines rather than just the yaw angle.

  author={Li, Yuchen and Li, Zixuan and Teng, Siyu and Zhang, Yu and Zhou, Yuhang and Zhu, Yuchang and Cao, Dongpu and Tian, Bin and Ai, Yunfeng and Xuanyuan, Zhe and Chen, Long},
  booktitle={2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, 
  title={AutoMine: An Unmanned Mine Dataset}, 

Contact Us

We are the AutoMine Group(AMG), if you are interested in our work and want to find an academic research position, please contact