method: Shopee MMU OCR2022-09-28
Authors: Jianqiang Liu, Hanfei Xu, Bin Zheng, Eric W, Ronnie T, Alex X
Affiliation: Shopee MMU
Description: Our method adopts a transformer-based context-aware framework. We utilize a hybrid architecture encoder and a context-aware autoregressive decoder to construct the recognition pipeline. Finally, a simple but effective multi-model fusion strategy is adopted.
method: SogouMM2019-11-07
Authors: Xu Liu, Tao Wei
Description: Our method is based on 2D-attention, we use ResNet as backbone and a tailored 2D-attention module is applied. The result is generated by single model without ensemble tricks.
method: Hancom Vision2020-10-06
Authors: Hancom Vision team
Description: Our model is featured by CNN-based, BiLSTM, and Attention.
Trained on MJSynthText + SynthText + external data (Pretrain), Focused Scene Text 2013-2015, and Incidental Scene Text 2015.
| Date | Method | Total Edit distance (case sensitive) | Correctly Recognised Words (case sensitive) | T.E.D. (case insensitive) | C.R.W. (case insensitive) | |||
|---|---|---|---|---|---|---|---|---|
| 2022-09-28 | Shopee MMU OCR | 134.8110 | 87.14% | 104.6682 | 89.17% | |||
| 2019-11-07 | SogouMM | 144.5029 | 86.42% | 113.1573 | 88.11% | |||
| 2020-10-06 | Hancom Vision | 160.2667 | 86.09% | 108.3773 | 88.93% | |||
| 2019-10-31 | Sogou_OCR | 163.0954 | 84.35% | 129.2831 | 86.66% | |||
| 2018-09-13 | Clova AI / Lens | 175.4367 | 83.00% | 132.4229 | 85.56% | |||
| 2020-06-10 | test 1 | 164.4290 | 82.91% | 129.2433 | 85.07% | |||
| 2018-07-03 | Baidu VIS | 185.8078 | 82.85% | 150.8527 | 84.68% | |||
| 2018-12-19 | SAR | 437.1642 | 67.40% | 203.1446 | 78.82% | |||
| 2016-01-29 | SRC-B-TextProcessingLab | 419.7412 | 62.11% | 367.1222 | 64.95% | |||
| 2015-11-09 | Megvii-Image++ | 508.8323 | 57.82% | 377.6521 | 63.99% | |||
| 2015-04-01 | MAPS | 1,128.0075 | 32.93% | 1,068.7184 | 33.90% | |||
| 2015-04-01 | NESP | 1,164.4968 | 31.68% | 1,094.7071 | 32.98% | |||
| 2015-04-02 | DSM | 1,178.6140 | 25.85% | 1,108.9381 | 27.97% |