Benchmark dataset for Earth monitoring

本论文提议开发一个针对地球监测的基础模型基准数据集，专注于气候变化相关的下游任务。该数据集旨在整合多种任务，如分类、语义分割等，以评估基础模型在处理不同类型的气候相关任务时的性能。数据集的设计考虑了多个方面，包括数据大小、许可、多光谱和SAR数据的包含，以及元信息以评估分布变化。此基准数据集的目标是促进地球监测领域基础模型的发展，提供一个系统的方法来衡量模型质量，并预先减少基础模型可能产生的负面影响。

This paper proposes the development of a foundation model benchmark dataset for Earth monitoring, focusing on downstream tasks related to climate change. This dataset is designed to integrate diverse tasks including classification, semantic segmentation, and more, to evaluate the performance of foundation models on various climate-related tasks. The dataset’s design takes into account multiple aspects, including data scale, licensing terms, inclusion of multispectral and SAR data, and meta-information for assessing distribution shifts. The goal of this benchmark dataset is to facilitate the development of foundation models in the field of Earth monitoring, provide a systematic approach to assess model quality, and preemptively reduce the potential negative impacts that foundation models may generate.