method: PingAn_VisualComputing2020-10-20

Authors: Dong Mei (梅栋), Xianbiao Qi (齐宪标), Yihao Chen (陈意浩), Di Wu (吴迪), Rong Xiao (肖嵘)

Affiliation: Ping An Property & Casualty Insurance Co of China Ltd. (中国平安财产保险股份有限公司)

Description: Our method ensembles multiple MASTER [1] and SAR [2] models. Our training datasets consist of several public datasets, including ReCTS, Art, CTW, RCTW, LSVT, MLT，Baidu Scene Text Recognition contest data (百度中文场景文字识别大赛), and some synthesized data based on CTW and ReCTS.
Our implementation is based on our own toolbox, FastOCR. FastOCR is a simple, fast but powerful text detection and recognition toolbox. We implement CRNN, SAR, MASTER, EAST, PSENet, Mask R-CNN, and several other SOTA methods in FastOCR.

1. Lu, N., Yu, W., Qi, X., Chen, Y., Gong, P., & Xiao, R. (2019). Master: Multi-aspect non-local network for scene text recognition. arXiv preprint arXiv:1910.02562.

2. Li, H., Wang, P., Shen, C., & Zhang, G. (2019, July). Show, attend and read: A simple and strong baseline for irregular text recognition. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 33, pp. 8610-8617).

method: TPS-ResNet v12019-04-30

Authors: Jeonghun Baek, Moonbin Yim, Sungrae Park, and Hwalsuk Lee

Description: We used Thin-plate-spline (TPS) based Spatial transformer network (STN) which normalizes the input text images, ResNet based feature extractor, BiLSTM, and attention mechanism.
This model was developed based on the analysis of scene text recognition modules.
See our paper and source code.

[Training Data]
At first, we generated the Chinese synthetic datasets by MJSynth and SynthText code, then pre-trained our model with the synthetic dataset and real dataset (ArT, LSVT, ReCTS, and RCTW). After that, we finetuned it with ReCTS data.

What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis (Accepted by ICCV 2019 oral)

Source code

method: HUST_Reg2019-04-30

Authors: Qingquan Xu (徐清泉), Wuheng Xu(徐武恒)，Qiuhui Huang(黄秋慧), Zhijun Xue(薛智钧), Changxu Cheng(程昌旭)

Description: Reference to the paper "ASTER: An Attentional Scene Text Recognizer with Flexible Rectification" and "AON: Towards Arbitrarily-Oriented Text Recognition".
Training datasets: ReCTS, MTWI 2018, ART.
Organization: Huazhong University Of Science And Technology(HUST，华中科技大学)
Email: qingquanxu@hust.edu.cn

ASTER: An Attentional Scene Text Recognizer with Flexible Rectification

AON: Towards Arbitrarily-Oriented Text Recognition

Ranking Table

Description Paper Source Code
DateMethodResult
2020-10-20PingAn_VisualComputing96.62%
2019-04-30TPS-ResNet v194.77%
2019-04-30HUST_Reg89.01%
2019-04-27resnet101lstm78.26%
2019-04-30baseline75.50%

Ranking Graphic