Aerial Image Patches Dataset

This dataset consists of 6283 color aerial image patches with a resolution of 512x512 pixels. The patches were extracted from orthophoto imagery obtained from the National Geoportal service and the USGS National Agricultural Imagery Program (NAIP). Both sources provide data in the public domain.The dataset was originally prepared for training deep learning models for automatic image colorization, but it can also be used for other computer vision tasks such as segmentation or classification. The collected patches cover a variety of landscapes including urban areas, agricultural fields, forests, and water bodies. Images depicting winter conditions or high levels of cloud coverage were excluded so that the dataset better reflects the conditions under which historical aerial photographs were taken. Exemplary images from the dataset The dataset is organized into two directories: train and val. Each directory contains the corresponding subset of images in PNG format. The training subset contains the majority of the data, while the validation subset is intended for model evaluation during training.