SpaceNet 7 (Multi-Temporal Urban Development SpaceNet Dataset)

卫星图像分析有许多人类发展和灾害响应应用，特别是在涉及时间序列方法时。例如，量化人口统计是 232 个联合国可持续发展目标中的 67 个的基础，但世界银行估计，目前有 100 多个国家缺乏有效的民事登记系统。 SpaceNet 7 Multi-Temporal Urban Development Challenge 旨在帮助解决这一缺陷并为非视频时间序列数据开发新的计算机视觉方法。在这项挑战中，参与者将在快速城市化地区收集的卫星图像时间序列中识别和跟踪建筑物。比赛围绕一个新的 Planet 卫星图像马赛克开源数据集展开，其中包括 24 张图像（每月一张），涵盖约 100 个独特的地理区域。该数据集将包含超过 40,000 平方公里的图像和图像中建筑物足迹的详尽多边形标签，总计超过 1000 万个单独的注释。挑战参与者将被要求随着时间的推移跟踪建筑施工，从而直接评估城市化。Satellite image analysis has numerous applications for human development and disaster response, particularly when time-series methods are employed. For instance, the quantification of population statistics underpins 67 of the 232 United Nations Sustainable Development Goals, yet the World Bank estimates that over 100 countries currently lack effective civil registration systems. The SpaceNet 7 Multi-Temporal Urban Development Challenge aims to help address this gap and develop novel computer vision methods for non-video time-series data. Within this challenge, participants will identify and track buildings in time-series satellite imagery collected from rapidly urbanizing regions. The competition centers on a new open-source dataset of Planet satellite image mosaics, consisting of 24 monthly images spanning approximately 100 unique geographic regions. This dataset will cover over 40,000 square kilometers of imagery, paired with exhaustive polygonal labels for building footprints within the scenes, totaling more than 10 million individual annotations. Challenge participants will be required to track building construction over time to directly evaluate urbanization.

搜集汇总

背景与挑战

背景概述

SpaceNet 7是一个多时序城市发展卫星图像数据集，专注于二维语义分割任务，旨在通过时间序列卫星图像分析支持城市化评估和人类发展应用。数据集包含约100个地理区域的24张月度Planet卫星图像马赛克，覆盖超过40,000平方公里，并提供超过1000万个建筑物足迹的多边形标签，用于跟踪建筑施工和城市化进程。该数据集由亚马逊、Planet和Capella Space于2021年发布，采用CC BY-SA 4.0许可协议，适用于计算机视觉研究，特别是非视频时间序列数据分析。