Multiple Distribution Shift - Aerial (MDS-A)
收藏资源简介:
MDS-A数据集是由亚利桑那州立大学和阿根廷国家南部大学计算机科学与工程系共同创建的一组模拟空域数据集,旨在研究天气条件变化对目标检测模型性能的影响。该数据集包含了6种不同天气条件下模拟的空中影像,每种天气条件下的训练集包含1000张图像,以及对应的测试集。数据集中的图像是在AirSim模拟器中捕获的,并且为每个场景中的物体提供了边界框标注。该数据集可用于评估模型在分布偏移方面的鲁棒性,并提供了基线模型和错误检测规则的结果。
The MDS-A dataset is a simulated airspace dataset jointly created by Arizona State University and the Department of Computer Science and Engineering, National University of the South. It aims to investigate the impact of varying weather conditions on the performance of object detection models. This dataset includes simulated aerial imagery under six distinct weather conditions. For each weather condition, the training set contains 1000 images along with a corresponding test set. All images in the dataset are captured using the AirSim simulator, and bounding box annotations are provided for objects in each scene. This dataset can be used to evaluate the robustness of models against distribution shifts, and it also provides the results of baseline models and error detection rules.




