University-1652
收藏资源简介:
该数据集是一个大规模的多视角多源无人机地理定位基准,涵盖了来自仿真无人机、卫星和地面相机的图像。它包含了来自全球72所大学的1,652栋建筑物,旨在提高无人机视角与卫星视角图像之间的匹配准确性。此外,该数据集还包括了701栋建筑物,训练集和测试集之间没有重叠,并通过图像增强技术缓解了不同领域图像的不平衡问题。该数据集规模宏大,拥有50,195张训练图像和37,855张测试图像,其任务重点是无人机视角目标定位和无人机导航。
This dataset is a large-scale multi-view and multi-source drone geolocalization benchmark, covering images captured by simulated drones, satellites, and ground cameras. It includes 1,652 buildings from 72 universities worldwide, aiming to improve the matching accuracy between drone-view and satellite-view images. Additionally, it contains another 701 buildings, with no overlap between the training and test sets. Data augmentation techniques are adopted to alleviate the data imbalance issue across different domains. Boasting a substantial scale, the dataset has 50,195 training images and 37,855 test images, whose core tasks focus on drone-view target geolocalization and drone navigation.




