CODrone
收藏资源简介:
CODrone 是一个为无人机设计的全面定向目标检测数据集,它准确反映了真实世界条件。该数据集包含来自多个城市在不同光照条件下的广泛标注图像,增强了基准的逼真度。CODrone 包含超过 10,000 张高分辨率图像,捕获自五个城市的真实无人机飞行,涵盖了各种城市和工业环境,包括港口和码头。为了提高鲁棒性和泛化能力,它包括在正常光线、低光和夜间条件下相同场景的图像。我们采用了三种飞行高度和两种常用的相机角度,从而产生了六个不同的视角配置。所有图像都针对 12 个常见对象类别进行了定向边界框标注,总计超过 590,000 个标记实例。总体而言,这项工作构建了一个综合数据集和基准,用于城市无人机场景中的定向目标检测,旨在满足该领域的研究和实践应用需求。
CODrone is a comprehensive oriented object detection dataset designed for drones, which accurately reflects real-world conditions. This dataset contains extensively annotated images collected from multiple cities under varying lighting conditions, enhancing the realism of the benchmark. CODrone includes over 10,000 high-resolution images captured during real drone flights across five cities, covering various urban and industrial environments including ports and docks. To improve robustness and generalization capability, it includes images of the same scenes under normal light, low light, and nighttime conditions. We adopted three flight altitudes and two commonly used camera angles, resulting in six distinct viewpoint configurations. All images are annotated with oriented bounding boxes for 12 common object categories, totaling more than 590,000 labeled instances. Overall, this work constructs a comprehensive dataset and benchmark for oriented object detection in urban drone scenarios, aiming to meet the research and practical application needs of this field.
CODrone数据集概述
数据集基本信息
- 名称: CODrone (Comprehensive Oriented Object Detection benchmark for UAV)
- 类型: 无人机(UAV)定向目标检测数据集
- 规模:
- 图像数量: 10,000张
- 标注对象: 596,700个
- 分辨率: 3840×2160 (4K)
- 标注格式:
- DOTA格式
- VOC格式
- 标注类型: 定向边界框(OBB)
关键特性
- 多样性:
- 12个对象类别
- 多种场景: 城市、乡村、港口、工业区等
- 飞行条件多样性:
- 飞行高度: 30m, 60m, 100m (共70m范围)
- 相机角度: 30°和90°
- 技术特点:
- 高分辨率图像采集
- 精确的定向边界框标注
数据集结构
CODrone/ ├── test/ │ ├── annfile/ # DOTA格式标注 │ ├── images/ │ └── labels/ # VOC格式标注 ├── train/ │ ├── annfile/ │ ├── images/ │ └── labels/ └── val/ ├── annfile/ ├── images/ └── labels/
对比优势
| 特性 | CODrone | 其他数据集(平均) |
|---|---|---|
| 分辨率 | 3840×2160 | 1680×1024 |
| 类别数量 | 12 | 5 |
| 飞行高度范围 | 70m | 44m |
| 相机角度 | 60° | 37.5° |
| OBB支持 | ✅ | 部分支持 |
许可信息
- 许可证: CC BY-NC-SA 4.0
- 使用限制: 仅限学术研究使用
获取方式
- Google Drive: https://drive.google.com/file/d/1vZb8VKNs-bFZNB68ar1WR3wPWyL824Ha/view?usp=sharing
- Quark Drive: https://pan.quark.cn/s/d65dd2124a04
引用格式
bibtex @misc{ye2025clearflexibleprecisecomprehensive, title={More Clear, More Flexible, More Precise: A Comprehensive Oriented Object Detection benchmark for UAV}, author={Kai Ye and Haidi Tang and Bowen Liu and Pingyang Dai and Liujuan Cao and Rongrong Ji}, year={2025}, eprint={2504.20032}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2504.20032}, }
联系方式
- 邮箱: yekai@stu.xmu.edu.cn




