Major-TOM/Core-S2RGB-SigLIP

Core-S2RGB-SigLIP数据集包含Sentinel-2 Level 2A (RGB)卫星图像的嵌入表示，共有20,212,974个嵌入，大小为41.3 GB。嵌入是通过SigLIP模型的图像编码器提取的，该模型是一个视觉-语言模型。数据集的主要目的是提供一种标准化的方式来发布Major TOM数据集的嵌入扩展，以便在减少存储和计算需求的情况下浏览和导航大型数据集。数据集的内容包括唯一ID、嵌入数组、网格单元、产品ID、时间戳、中心经纬度、几何形状等信息。数据集是由CloudFerro和欧洲空间局的Φ-lab合作开发的，并在CREODIAS云服务平台上使用GPU加速实例计算。

The Core-S2RGB-SigLIP dataset contains embeddings of Sentinel-2 Level 2A (RGB) satellite imagery, comprising 20,212,974 embeddings with a size of 41.3 GB. The embeddings were extracted using the image encoder of the SigLIP model, a vision-language model. The primary purpose of the dataset is to provide a standardized way to release embedding expansions of the Major TOM datasets, enabling the browsing and navigation of large datasets with reduced storage and computational demands. The dataset includes fields such as unique_id, embedding array, grid_cell, product_id, timestamp, centre_lat, centre_lon, geometry, and more. The dataset was developed in collaboration between CloudFerro and the Φ-lab of the European Space Agency, and was computed on GPU-accelerated instances on the CREODIAS cloud service platform.