"Building detection using aerial imagery in poorly planned regions: the case of San Jos\u00e9 de las Matas (Dominican Republic)"

"The Building Detection Using Aerial Imagery in Poorly Planned Regions: the Case of San Jos\u00e9 de las Matas (Dominican Republic) dataset is a high-resolution, manually annotated aerial image collection designed to support research in automated building segmentation under irregular urban conditions. Capturing a 3.41 km\u00b2 area of San Jos\u00e9 de las Matas\u2014a municipality marked by informal construction, low-rise residential structures, and occlusion from dense vegetation\u2014the dataset addresses a critical gap in training and evaluating deep learning models beyond well-planned cityscapes. It includes 112 RGB aerial images of size 640\u00d7640 pixels, each paired with a pixel-level building mask. The imagery, sourced from Google Maps, was labeled using the Labelbox platform to ensure accurate delineation of building footprints despite occlusions and irregular layouts. This dataset poses unique challenges, such as varied roof geometries, unaligned building orientations, and frequent vegetation overlap\u2014features rarely present in conventional benchmarks. It serves as a valuable resource for testing the generalization and transferability of convolutional neural networks, particularly U-Net-based architectures trained on structured datasets. By enabling experimentation in data-scarce and underrepresented regions, this dataset contributes to more inclusive and scalable solutions for urban mapping and population estimation in the Global South.   "