method: Ping An Property & Casualty Insurance Company2019-04-22

Authors: Xianbiao Qi, Ning Lu, Yuan Gao, Yihao Chen, Shaoqiong Chen, Wenwen Yu, Rong Xiao

Description: We use an anchor-free detection framework with FishNet as our backbone. We pre-process the training and testing images to roughly align their scales with OpenCV adaptive threshold. We post-process the detection results according to the classification score map. It can well solve the problem of short detection of long text line.

1. Sun, Shuyang, et al. "Fishnet: A versatile backbone for image, region, and pixel level prediction." Advances in Neural Information Processing Systems. 2018. 2. Zhou, Xinyu, et al. "EAST: an efficient and accurate scene text detector." Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. 2017. 3. Zhang, Zhi, et al. "Bag of Freebies for Training Object Detection Neural Networks." arXiv preprint arXiv:1902.04103 (2019).