UAV-VisLoc: A Large-scale UAV Dataset with Continuous Trajectories for Visual Geo-localization

Visual geo-localization of unmanned aerial vehicles (UAVs) in GNSS-denied environments remains challenging due to the lack of datasets with trajectory continuity, broad scene and geographic coverage, multimodal UAV imagery, wide altitude and geometric variation, and sufficient scale. Here we present UAV-VisLoc, a dataset for UAV geo-localization through image retrieval between UAV imagery and geo-referenced satellite tiles. UAV-VisLoc contains 14,998 UAV images organized into 30 flight trajectories, spanning six continents and seven scene types. The dataset preserves contiguous observations along each trajectory, enabling trajectory-level studies under realistic sequential flight conditions. It includes RGB and Infrared UAV imagery and covers altitudes from 160 to 2,500 m. The dataset comprises 21 real UAV flights collected across
multiple provinces in China and 9 synthetic flights built from OpenAerialMap imagery for regions outside China. The dataset also provides image-level flight metadata and more than 120,000 geo-referenced satellite tiles. Baseline retrieval experiments and accompanying scripts are provided to validate data organization, ground-truth annotations, and benchmark usability under diverse real-world conditions.