SpaceNet 2 (SpaceNet 2: Building Detection v2)

SpaceNet 2: Building Detection v2 - 是一个数据集，用于根据非常高分辨率的卫星图像在地理不同的环境中构建足迹检测。它包含超过 302,701 个建筑足迹，3/8 波段 Worldview-3 卫星图像，分辨率为 0.30 万像素，横跨 5 个城市（里约热内卢、拉斯维加斯、巴黎、上海、喀土穆），涵盖城市和郊区的性质。使用 60%/20%/20% 分割数据集用于训练/测试/验证。从卫星图像中检测建筑物足迹的主要用例是辅助基础测绘。

SpaceNet 2: Building Detection v2 is a dataset dedicated to building footprint detection across geographically diverse environments using very high-resolution satellite imagery. It comprises over 302,701 building footprints, paired with 3/8-band Worldview-3 satellite imagery at a ground sampling distance of 0.30 meters per pixel, spanning five cities including Rio de Janeiro, Las Vegas, Paris, Shanghai and Khartoum, and covering both urban and suburban settings. The dataset is split at a 60%/20%/20% ratio for training, testing and validation, respectively. The primary use case of building footprint detection from satellite imagery is to assist with basic base mapping.

搜集汇总

背景与挑战

背景概述

SpaceNet 2是一个用于建筑物检测的卫星图像数据集，包含超过30万个建筑足迹，覆盖全球5个地理多样性城市，图像分辨率为0.30万像素。该数据集专为训练和评估计算机视觉模型设计，支持语义分割、实例分割和物体检测任务，主要用于辅助基础测绘应用。