rootsautomation/GloSAT

GloSAT表格数据集是一个用于表格检测和表格结构识别的数据集，专门针对扫描的历史气象日志页面。该数据集包含来自9个不同来源的499张图像，涵盖19世纪和20世纪初的档案资料。每张图像都标注了表格边界框以及每个单元格的边界框，包括行/列索引和标题标志。数据集提供两个配置：coarse配置将合并单元格视为单个单元，平均每个表格有156个单元格；fine配置为每个物理单元格单独标注，平均每个表格有488个单元格。数据集不包含转录的单元格内容，仅提供空间布局信息（如边界框和网格拓扑）。数据分为371张训练图像和129张测试图像，适用于历史文档图像处理任务。

The GloSAT Table Dataset provides table structure recognition annotations for scanned historical meteorological logbook pages from the GloSAT project. Images are scanned pages from 9 archival sources spanning the 19th and early 20th centuries. Each image is annotated with table bounding boxes and per-cell bounding boxes with row/column indices and header flags. The dataset supports table detection and table structure recognition (TSR) tasks on historical document images. It includes two configs: coarse treats merged cells as single units, while fine annotates each physical cell individually. No transcribed cell content is included; it is a structure-only dataset providing spatial layout. The dataset contains 499 images split into 371 train and 129 test images.