HR-VSGeo: A High-Resolution Visible-SAR Geo-Referenced Dataset and Open Evaluation Benchmark for Disaster-Prone Areas

We present HR-VSGeo, a high-resolution, geo-referenced Visible\u2013SAR image dataset with an open evaluation benchmark for disaster-prone regions. HR-VSGeo comprises 5,147 optical images and 1,319 pairs of co-registered optical\u2013SAR images spanning Southeast Asia and southwestern China. It supports multiple tasks\u2014including object mapping and recognition in disaster-prone contexts, semantic segmentation, and multi-modal fusion. The dataset provides more than 110,000 annotations across 15 disaster-related categories (e.g., buildings, roads). All image tiles and instances are geo-referenced and labeled with bounding boxes and minimum bounding polygons, enabling precise localization and spatial analysis.