LocBench

LocBench contains 7 geo-aware image classification datasets and 4 geo-aware image regression datasets, aiming to systematically compare the location encoders' performance and their impact on the model's overall geographic bias.
Geo-aware image classification datasets: BirdSnap, BirdSnap†, NABirds†, iNat2017, iNat2018, YFCC, and fMoW. Please note that BirdSnap and BirdSnap† are in the same file birdsnap.tar.gz. iNat2017 and iNat2018 are not stored here due to storage limitations but can be accessed via Dropbox.Geo-aware image regression datasets: MOSAIKS (Population Density, Forest Cover, Nightlight Luminosity, and Elevation), and SustainBench (Asset Index, Women BMI, Water Index, Child Mortality Rate, Sanitation Index, and Women Edu). The SustainBench datasets can be downloaded from Google Drive. You can find dhs_test_labels.csv and dhs_trainval_labels.csv in sustainbench_dhs_labels.zip here if they were not in the Google Drive.All data products created through our work that are not covered under upstream licensing agreements are available via a CC BY-NC 4.0 license. All upstream data use restrictions take precedence over this license.