OCTT: Occluded Character-level Total-Text Dataset

OCTT (Occluded Character-level Total-Text) is a dataset for benchmarking scene text detection, recognition, and spotting under occlusion. It extends the Total-Text dataset with synthetically occluded versions of scene images (detection/) and cropped word instances (recognition/total-text-cropped/), each paired with the corresponding original (unoccluded) version. It was introduced in the paper "Occluded Text Detection and Recognition in the Wild" (Raisi & Zelek, CRV 2022). Code: https://github.com/zobeirraisi/occluded_text_detection_recognitin