CRASAR-U-DROIDs

This dataset contains 52 orthomosaics containing 21716 building polygons collected from 10 different disasters, totaling 67 gigapixels of imagery. Building polygons were sourced from Microsoft's US Building Footprint's Dataset, and in some cases building polygons were added manually by the authors. Each building polygon has been annotated using the Joint Damage Scale and translationally aligned for model training. The dataset has been split into test and train at the disaster level with 6 disasters in the train set, and 4 disasters in the test set. A summary of the dataset, grouped by disaster and ordered by area, is included below for reference.

本数据集包含52幅正射影像镶嵌图（orthomosaics），涵盖从10起不同灾害事件中采集得到的21716个建筑多边形（building polygons），总影像像素规模达67吉像素（gigapixels）。建筑多边形数据源自微软美国建筑足迹数据集（Microsoft US Building Footprint Dataset），部分建筑多边形由本文作者手动补充添加。每个建筑多边形均采用联合损伤等级（Joint Damage Scale）进行标注，并完成平移配准以适配模型训练。数据集按照灾害场景进行训练集与测试集划分：训练集包含6起灾害的样本，测试集包含4起灾害的样本。下文附带按灾害分组、按面积排序的数据集汇总信息，以供参考。

搜集汇总

背景与挑战

背景概述

CRASAR-U-DROIDs数据集包含来自10种不同灾害的52个正射影像，共计21716个建筑多边形和67千兆像素的影像数据。数据集已按灾害级别分为训练集和测试集，其中6个灾害用于训练，4个用于测试，所有建筑多边形均使用联合损伤量表进行注释并进行平移对齐，适用于模型训练。