method: TextMask_V12019-05-01

Authors: LiLin, Binbin Xu

Description: We propose a novel text detector called TextMask, a new Mask R-CNN based
text detection approach which can robustly detect irregular text from natural scene images. Our approach formulates arbitrary text detection as a instance segmentation problem and employ an FCN to make a pixel-level text/non-text prediction, based on which a text saliency map can be generated.