KATE-CD: high-resolution change detection dataset from the 2023 Earthquake in Kahramanmaraş, Türkiye

Earthquake damage assessment using bitemporal remote sensing images plays a crucial role in supporting disaster response and recovery efforts. With the growing availability of high-resolution remote sensing imagery and the advancement of change detection models, detailed damage maps can be generated to identify damaged buildings, obstructed roads, disrupted power grids, and other critical infrastructure failures. However, unlike many computer vision datasets, there is a significant lack of publicly available and task-specific datasets for earthquake damage assessment, which limits the development and benchmarking of advanced models. To address this gap, we introduce Kahramanmaraş Türkiye Earthquake-Change Detection (KATE-CD), a multisource, high-resolution, and manually annotated public dataset, curated specifically for change detection in earthquake-affected regions. Derived from the 2023 Kahramanmaraş earthquakes in Türkiye, KATE-CD includes pre- and post-event satellite imagery and carefully drawn damage polygons, enabling fine-grained semantic segmentation of damaged structures in complex urban environments. In addition, we provide a fully reproducible benchmark pipeline built upon the SAM-CD framework, which incorporates segment anything models (SAMs) and a semantic learning module to extract latent representations from bitemporal imagery. To our knowledge, this is the first earthquake damage dataset from Türkiye paired with a benchmark that systematically evaluates state-of-the-art vision foundation models. We evaluate three vision encoders—ResNet, Fast SAM, and Efficient SAM—within the SAM-CD framework and report their performance across standard metrics. We also tested a vanilla UNet model to evaluate the merit of using foundation models. By combining a unique, high-impact dataset with a transparent evaluation protocol, we offer the research community a reliable and scalable foundation for developing next-generation post-disaster damage assessment models.