method: SROIE Fourth Submission2019-04-23

Authors: yangliu

Description: This is a multi-oriented scene text detection method based image segmentation. And it is a two-step method. The first step is to get the feature maps through the network and generate candidate bounding boxes. The second step is to filter the candidate text boxes through a two-stage NMS, so as to get the final results. What'more, we augmented and equalized the data. And the submission is by single-scale testing.