Major-TOM/Core-S2RGB-249k-DINOv2
收藏资源简介:
Core-S2RGB-249k-DINOv2是一个基于Core-S2L2A-249k数据集使用DINOv2-large模型计算的视觉嵌入数据集。数据集包含248,719个384×384像素的Sentinel-2 L2A图像块,经过预处理后生成768维的嵌入向量。输入波段为RGB(B04, B03, B02),输出格式为GeoParquet,嵌入维度为768。数据集的计算流程包括预处理、调整大小、编码、后处理和地理空间元数据保留。文件布局包括一个主要的嵌入GeoParquet文件和一个README文件。数据集的模式包括多个字段,如unique_id、embedding、timestamp等。
Core-S2RGB-249k-DINOv2 is a visual embedding dataset computed using the DINOv2-large model based on the Core-S2L2A-249k dataset. It contains 248,719 Sentinel-2 L2A image patches of 384×384 pixels, which generate 768-dimensional embedding vectors after preprocessing. Its input bands are RGB (B04, B03, B02), with an output format of GeoParquet and an embedding dimension of 768. The dataset's computation pipeline includes preprocessing, resizing, encoding, post-processing, and preservation of geospatial metadata. The file layout consists of a primary embedding GeoParquet file and a README file. The dataset schema includes multiple fields such as unique_id, embedding, timestamp, and others.




