TextSeg is a large-scale text detection and segmentation dataset with fine annotations and a wide range of uses.

The dataset includes six types of annotations: word-, character-wise bounding polygons, masks and transcriptions.