This is also an instance segmentation model, trained on extracting text-lines from the cropped text-regions. The same post-processing as in the text-region segmentation step, is done in the text-line segmentation step.
