S2-100K

该数据集包含了10万张多光谱的哨兵-2卫星图像瓦片及其相应的中心位置坐标，旨在适用于多任务应用和地理泛化性能的提升。该数据集旨在呈现普遍的地理位置特征，并且几乎均匀地分布在全世界的陆地上。规模达到了10万张瓦片。任务包括为各种下游任务预训练位置编码器，这些任务涵盖了温度预测、动物识别以及人口密度估计等。

This dataset contains 100,000 multispectral Sentinel-2 satellite image tiles along with their corresponding central geographic coordinates. It is developed to support multi-task applications and enhance geographic generalization performance. The dataset aims to capture universal geographic features and is almost uniformly distributed across global land areas. The supported tasks of this dataset include pre-training location encoders for various downstream tasks, which cover temperature prediction, animal recognition, population density estimation, and more.

搜集汇总

背景与挑战

背景概述

S2-100K数据集包含100,000张无云覆盖的多光谱卫星图像，采样自Sentinel-2数据，时间跨度为2021年1月至2023年5月，适用于研究用途。