RWanda Built-up Region Segmentation

我们创建了卢旺达建成区域数据集，它与以前可用的数据集在本质上不同且用途广泛。不同的结构大小和形态，不规则的建筑模式，森林和沙漠中的建筑，以及泥屋的存在，使其非常具有挑战性。以每像素 1.193 米的高分辨率 (HR) 收集了总共 787 张大小为 256 × 256 的卫星图像，并使用在线工具 Label-Box 为构建区域分割进行手工标记。We have developed a built-up area dataset for Rwanda, which is fundamentally distinct from previously available datasets and has a wide range of applications. Variations in structural sizes and morphologies, irregular building patterns, buildings situated within forests and deserts, as well as the presence of mud-brick dwellings, make this dataset particularly challenging. A total of 787 satellite images, each sized 256 × 256, were collected at a high resolution (HR) of 1.193 meters per pixel, and manually annotated for built-up area segmentation using the online tool Label-Box.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于建成区域分割的二维语义分割数据集，包含787张高分辨率卫星图像，每像素1.193米，图像大小为256×256，覆盖卢旺达地区，具有不规则建筑模式等挑战性特征。数据由Information Technology University于2020年发布，用于领域适应研究。