DocDewarpHV

This dataset contains about 110K distorted document images in Chinese and English. The number of Chinese and English documents is close to 1:1. The resolution of each image is 512*512. The source scanned images come from cddod, CDLA, M6Doc and PubLayNet. Compared with Doc3D, in addition to 3D world coordinates, UV map, 2D backward map (grid coordinates), we also provide horizontal and vertical line annotations that are consistent with the distortion trend of the input image.