Disaster

灾难是一个数据集，其中包含从各种来源收集的三种不同灾难的图像：火灾、水和土地。除此之外，它还包含各种因自然或人为灾难而受损的基础设施以及因战争或事故而受损的人类的图像。 该数据集中有 13,720 张手动标注的图像，每张图像由三个人标注。作者还为一组 200 个测试图像提供了用边界框手动注释的判别图像类信息。图像是从不同的新闻门户、社交媒体和其他研究人员提供的标准数据集中收集的。This is a disaster dataset comprising images of three distinct disaster categories collected from diverse sources: fire, water-related disasters, and land-related disasters. Additionally, it also contains images of infrastructure damaged by natural or anthropogenic disasters, as well as images of humans impacted by wars or accidents. This dataset includes 13,720 manually annotated images, with each image being annotated by three separate annotators. The authors further provide manually annotated discriminative image class information paired with bounding boxes for a set of 200 test images. All images were gathered from various news portals, social media platforms, and standard datasets supplied by other researchers.

搜集汇总

背景与挑战

背景概述

Disaster数据集包含13,720张手动标注的图像，涵盖火灾、水灾、土地灾害等灾难类型，以及受损基础设施和人类的图像。每张图像由三人标注，并提供200张测试图像的边界框注释，数据来源于新闻门户、社交媒体和其他研究数据集。该数据集由孟加拉国独立大学和达卡大学于2021年发布。