TabStructDB

In ICDAR-17, a Page-Object Detection (POD) competition was organized where the task was to identify page objects in documents which includes tables, figures and equations in document. The dataset was composed of 2417 images in total, where 1600 images were used for training, while the rest of the 817 images were used for testing. We are introducing a new table structure recognition dataset, TabStructDB, where we labeled each tabular region present in the ICDAR-17 POD dataset with table structure information comprising of the row and column information.