method: EAST_clip_enhance_896_giou2019-04-17

Authors: zhaopeng

Description: Based on the EAST scene text detection framework, the central area of the data is cropped and the contrast of fuzzy image is enhanced.Test image width all zoom to 896px, high scale zoom.EAST's bounding box loss function USES GIoU loss and replaces IoU loss, which improves the prediction accuracy of bounding box to a certain extent.

Zhou X, Yao C, Wen H, et al. EAST: an efficient and accurate scene text detector[C]//Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. 2017: 5551-5560.

Source code