Dual-temporal image-text dataset (DTITD130H)

DTITD130H is a dual-temporal image-text dataset that is used to improve the understanding of fine-grained geo-object relations for ORSI. As shown in Table A.I, DTITD130H comprises 13,000 ORSIs collected from different platforms, such as EarthExplorer, Google Earth and Esri World Imagery, as well as from visual tracking benchmarks such as OTB100 , RAIVD and SatSOT. To implement an efficient training process, large ORSIs are sliced into 256\u00d7256-pixel sub-images. DTITD130H includes 2,800 pieces of geospatial vector data collected from OpenStreetMap(OSM); 1,200 pieces of trajectory data drawn from OTB100, RAIVD and SatSOT; and 2,100 relation knowledge texts collected from open geographic knowledge bases and encyclopedic resources(e.g., Wikipedia).