Lexxurius/LMDB_Overhead_Geopose

该数据集是一个大规模、预处理过的数据集，用于从单张卫星RGB图像中进行人为物体的密集高度回归。数据集包含从卫星RGB图像中提取的256×256像素块及对应的地面以上高度（AGL）地图（DSM）。每个像素块包括一个二进制有效性掩码和捕获元数据，为地理空间计算机视觉中的密集回归任务提供了强大的训练支持。数据集采用Apache Parquet格式存储，支持批量读取而无需将整个数据集加载到RAM中。此外，数据集还提供了预计算的统计信息，包括区域和全局的均值/标准差，用于精确的RGB和高度归一化。This dataset is a large-scale, preprocessed dataset for dense height regression of anthropogenic objects from single satellite RGB imagery. It contains 256×256 patches extracted from satellite RGB imagery and corresponding Above-Ground Level (AGL) height maps (DSM). Each patch includes a binary validity mask and capture metadata, enabling robust training for dense regression tasks in geospatial computer vision. The dataset is stored in Apache Parquet format for batch-wise reading without loading the entire dataset into RAM. It also provides precomputed statistics, including regionally and globally pooled mean/std values for precise RGB and height normalization.