UCM Aerial Image Restoration Dataset

A paired dataset for all-in-one restoration of aerial images, built from the UC Merced Land Use dataset (Yang & Newsam, 2010). UC Merced is a widely used remote sensing benchmark of 2,100 aerial images covering 21 land-use classes, extracted from USGS National Map urban area imagery at 1-foot (0.3 m) resolution. From a class-balanced subset of 420 images (20 per class), this dataset provides 1,260 degraded–clean image pairs (1,008 train / 252 test) covering motion blur, Gaussian noise and atmospheric haze. All images are 256 × 256 PNG. If you use this dataset kindly cite the following papers, @article{NIJHUM2026117742, title = {Visual-conditioned prompting for all-in-one remote sensing image restoration}, journal = {Signal Processing: Image Communication}, pages = {117742}, year = {2026}, doi = {10.1016/j.image.2026.117742}, author = {Ifran Rahman Nijhum and Md Yearat Hossain and Md Shihab Reza and Md Mahbub Hasan Rakib and Soumyabrata Dev} } The original UC Merced dataset: @inproceedings{yang2010bag, author = {Yang, Yi and Newsam, Shawn}, title = {Bag-of-visual-words and spatial extensions for land-use classification}, booktitle = {Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems}, series = {GIS '10}, pages = {270--279}, year = {2010}, publisher = {Association for Computing Machinery}, address = {New York, NY, USA}, doi = {10.1145/1869790.1869829} }