5-class curated dataset for aerial military asset detection in UAV surveillance.
AMAD-5 是一个精选的无人机视角航拍数据集,包含五类军事与民用目标。该数据集专为航拍监控中的目标检测任务设计,涵盖车队、混合地形及平民场景等真实多目标环境。每张图像均提供 YOLO 格式 的边界框标注,可直接用于目标检测任务,也可适配图像分类任务。
AMAD-5 is a curated aerial dataset containing images of five classes of military and contextual assets captured from a UAV perspective. The dataset is designed for object detection in aerial surveillance scenarios and reflects real-world multi-object environments including convoys, mixed terrain, and civilian presence. Each image is annotated with bounding boxes in YOLO format, making the dataset directly usable for object detection tasks and adaptable for image classification if required.
| Class ID | Label | 说明 |
|---|---|---|
| 0 | military_tank | 军事坦克 |
| 1 | military_vehicle | 军用车辆(含多种轮式装备) |
| 2 | civilian | 平民 |
| 3 | soldier | 士兵 |
| 4 | civilian_vehicle | 民用车辆 |
military_vehicle类整合了多种轮式军事装备,确保航拍检测场景下的视觉一致性。
| Split | 图像数 / Images | 标签文件数 / Label Files | 大小 / Size |
|---|---|---|---|
| Train | 27,583 | 28,261 | 3,267.1 MB |
| Val | 4,184 | 4,268 | 451.5 MB |
| Test | 2,345 | 2,431 | 268.9 MB |
| Total | 34,112 | 34,960 | ~4,683 MB (4.7 GB) |
| 指标 / Metric | 数值 / Value |
|---|---|
| 总标注框 / Total Bounding Boxes | 70,704 |
| 每张图像目标数(最小) | 1 |
| 每张图像目标数(最大) | 86 |
| 每张图像目标数(平均) | 2.0 |
| 每张图像目标数(中位数) | 1 |
| Class ID | Label | 标注数 / Count | 占比 / Percentage |
|---|---|---|---|
| 0 | military_tank | 20,845 | 29.5% |
| 1 | military_vehicle | 15,873 | 22.4% |
| 2 | civilian | 13,372 | 18.9% |
| 3 | soldier | 11,773 | 16.7% |
| 4 | civilian_vehicle | 8,841 | 12.5% |
├── train/
│ ├── images/ # 27,583 张训练图像
│ └── labels/ # 28,261 个 YOLO 标注文件
├── val/
│ ├── images/ # 4,184 张验证图像
│ └── labels/ # 4,268 个 YOLO 标注文件
├── test/
│ ├── images/ # 2,345 张测试图像
│ └── labels/ # 2,431 个 YOLO 标注文件
├── dataset.yaml # 数据集配置文件
└── README.md所有标注文件遵循标准 YOLO 格式:
class_id x_center y_center width height免责声明: 本数据集仅用于学术研究目的,请遵守相关法律法规及数据使用条款。 Disclaimer: This dataset is intended for academic research purposes only. Please comply with applicable laws, regulations, and data usage terms.