method: H&H Lab2019-04-22

Authors: HUST_VLRGROUP(Hui Zhang, Mingkun Yang, Mengde Xu, Zhen Zhu, Jiehua Yang) & HUAWEI_CLOUD_EI(Jing Wang, Yibin Ye, Shenggao Zhu, Dandan Tu)

Description: We mainly completed the task 2 using CRNN. Different from the content in the paper, we modified the structure of CNN to PVANet-like and used multiple GRU layers. And we adjusted the training strategy to continue to improve the recognition result.

method: HeReceipt-Ensemble2019-04-22

Authors: Yichao Huang*, Tianwei Wang*, Jiaxin Zhang*, Yan Li, Jiapeng Wang, Canjie Luo, Kai Ding, Lianwen Jin (*equal contribution), INTSIG-AIM & SCUT-DLVC-Lab

Description: This is a recognition method based on CNN and RNN. With different backbones and recurrent structure settings, we train several models seperately and then implement model ensemble. Finally, according to the official requirement, we split line output into word level.

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

Description: We employ an encoder-decoder sequence method with attention mechanism. First, we create 2 millions of systhesis text lines, where the receipt background is used. Each line consists of one to five words. Then, we finetune the network with real-world receipt data.

1. Li, Hui, et al. "Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition." arXiv preprint arXiv:1811.00751 (2018). 2. Shi, Baoguang, et al. "Aster: An attentional scene text recognizer with flexible rectification." IEEE transactions on pattern analysis and machine intelligence (2018). 3. Vaswani, Ashish, et al. "Attention is all you need." Advances in neural information processing systems. 2017.

Ranking Table

Description Paper Source Code
DateMethodRecallPrecisionHmean
2019-04-22H&H Lab96.35%96.52%96.43%
2019-04-22HeReceipt-Ensemble94.56%95.10%94.82%
2019-04-22Ping An Property & Casualty Insurance Company94.48%94.86%94.67%
2019-04-22CLOVA OCR94.30%94.88%94.59%
2019-04-22SCUT-DLVC-Lab-Lexicon94.18%94.88%94.53%
2019-04-22DenseNet-Attention Recognition94.29%94.58%94.44%
2019-04-22CITlab Argus Text Recognition93.55%93.61%93.58%
2019-04-18Unet followed by CRNN with CTC88.58%87.30%87.93%
2019-04-23BOE_IOT_AIBD87.84%86.66%87.24%
2019-04-22CRNN after UNet Segmentation85.77%86.48%86.12%
2019-04-22BiLstmCtcIgnoreSpaces-Segment83.38%87.37%85.33%
2019-04-22IFLYTEK-textRec_v480.63%81.72%81.17%
2019-04-22SituTech_OCR77.81%80.10%78.94%
2019-04-19Attention based OCR70.87%72.27%71.57%
2019-04-18CTPN_CRNN35.75%63.89%45.85%
2019-04-22A Text Extraction Method Based on Modified CRNN26.33%72.53%38.63%
2019-04-18BiLSTM+ctc28.75%49.69%36.42%
2019-04-23CRNN_Pytorch_BiCTC21.43%45.65%29.16%
2019-04-21ICA-IVA19.48%40.32%26.27%
2019-04-22Task 2 - Scanned Receipt OCR (Submitted by Intuit Inc.)1.05%3.96%1.66%
2019-04-22Receipt Info Extracting Task2 letter-recognizing0.24%0.97%0.38%
2019-04-16VIL0.00%0.00%0.00%
2019-04-22CRNN_Autohome0.00%0.00%0.00%

Ranking Graphic