method: Tencent TEG OCR2020-03-15

Authors: Pei Xu, Shan Huang, Shen Huang, Qi Ju.

Description: We reimplemented the standalone recognition method according to the end-to-end text spotting code released by the Mask TextSpotter[TPAMI]. It is a seq-to-seq method based on 2D attention. We synthesize curved text images for pretraining by the method of VGG synthtext. We add public dataset including icdar2013-2015, CUTE, SVT, IIIT5k, RCTW2017, LSVT,ctw to finetune and don't use any private data.

Authors: Jeonghun Baek, Moonbin Yim, Junyeop Lee, and Hwalsuk Lee

Description: Before text recognition, we used the text detector called CRAFT as a preprocessing step.
For a recognition model, 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.

method: TPS-ResNet2019-04-30

Authors: Jeonghun Baek, Moonbin Yim, Junyeop Lee, 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.

Ranking Table

Description Paper Source Code
DateMethodResultTotal wordsCorrect words
2020-03-15Tencent TEG OCR85.74%4842635011
2019-05-01CRAFT (Preprocessing) + TPS-ResNet85.32%4842634206
2019-04-30TPS-ResNet83.63%4842633173
2019-04-30PKU Team Zero65.06%4842626216
2019-04-29NPU-ASGO63.82%4842625341
2019-04-21Arbitrary shape scene text recognition based on CNN and Attention Enhanced Bi-directional LSTM54.49%4842619792

Ranking Graphic