DOTA

该数据集名为DOTA，包含了大约2800张来自谷歌地球及特定卫星的图片，标注了包括飞机、船只和车辆在内的16个对象类别。为了便于对象检测模型的统一处理，这些图片被分割成了1024x1024像素的小块，并分别设置了训练集和测试集。该数据集的规模为2800张图片，任务是对对象进行检测，并评估对抗性补丁配置的效果。

The dataset named DOTA contains approximately 2,800 images collected from Google Earth and specific satellites, with annotations for 16 object categories including airplanes, ships, and vehicles. To enable unified processing by object detection models, these images were cropped into 1024×1024 pixel patches, and separate training and test sets were configured. The dataset, which totals 2,800 images, includes two core tasks: object detection and the assessment of adversarial patch configuration efficacy.