D’RespNeT - Instance Segmentation.v7i.yolov8.zip

Project DescriptionD’RespNeT is a high-resolution UAV-based dataset designed to support post-earthquake disaster response through instance segmentation of access points and obstacles. Unlike existing datasets that rely mainly on satellite imagery, D’RespNeT provides fine-grained polygon-level annotations across 28 critical categories, including damaged buildings, accessible and blocked entry points, debris levels, vehicles, civilians, and rescue personnel. Curated from 1080p aerial footage of recent disasters (e.g., the 2023 Türkiye–Syria earthquake), the dataset enables real-time situational awareness for search-and-rescue teams and robotic platforms (UAVs/UGVs). Benchmarked with state-of-the-art YOLO-based models, D’RespNeT demonstrates significant improvements in accuracy, boundary precision, and operational decision-making, making it a valuable tool for both research and practical deployment in disaster response.If you wish to cite or use this dataset in your research, please contact us directly. You can check the technical paper of this research from here: https://www.arxiv.org/abs/2508.16016<br>