- Task 1 - Character Recognition
- Task 2 - Text Line Recognition
- Task 3 - Text Line Detection
- Task 4 - End-to-End Text Spotting
method: CPN (multi-scale)2024-05-30
Authors: Longhuang Wu, Shangxuan Tian, Youxin Wang, Pengfei Xiong
Email: wlonghuang@gmail.com
Description: We propose a Complementary Proposal Network (CPN) that seamlessly and parallelly integrates semantic and geometric information for superior performance. This Result is achieved with single Swin-L backbone and multi-scale testing policy. No model ensemble is used.
method: CBL_OCR2022-01-30
Authors: Jingyi Shen(沈静逸), Guokun Wang(王国坤), Yue Wu(吴岳), Chang Zhou(周昌), Jianqiang Huang(黄建强)
Affiliation: Alibaba
Description: Anchor-free based detection framework with poly regression and text segmentation was used here. The model was firstly pretrained on LSVT、MLT、ReCTS,and then finetuned on ReCTS. Weighted loss was used to enhance small text instances. At testing phase, single-model-multi-scale with post-processing was used to generate the final results.
method: Unis_OCR2021-04-27
Authors: BaoLin.Zhang(张保林),Jie.Li(李杰),KeJie.Liu(刘克捷),Yuan.Hu(胡源)
Affiliation: UNISINSIGHT
Description: First, the training method is based on SAST framework, which takes RES101 with multi-scale features as the backbone, and adds attention and lamdba method . At the same time, the training data used include LSVT,LSVT Weak, ReCTS, Weak, icdar and other free data from internet.
Date | Method | Recall | Precision | Hmean | |||
---|---|---|---|---|---|---|---|
2024-05-30 | CPN (multi-scale) | 94.50% | 94.61% | 94.56% | |||
2022-01-30 | CBL_OCR | 93.27% | 94.08% | 93.67% | |||
2021-04-27 | Unis_OCR | 94.66% | 92.51% | 93.57% | |||
2019-10-10 | NSTD-iFLYTEK | 93.17% | 93.62% | 93.40% | |||
2019-05-01 | SANHL_v4 | 93.97% | 92.76% | 93.36% | |||
2019-05-01 | Tencent-DPPR Team | 93.46% | 92.59% | 93.03% | |||
2019-04-30 | maskrcnn_text | 91.96% | 90.09% | 91.02% | |||
2019-04-30 | Task3-re5 | 90.03% | 91.65% | 90.83% | |||
2019-04-23 | A region proposal and fcn model based method | 88.64% | 92.72% | 90.64% | |||
2019-04-30 | pursuer | 86.13% | 92.72% | 89.31% | |||
2019-04-29 | CLTDR | 88.92% | 88.70% | 88.81% | |||
2020-05-18 | MMTD | 86.63% | 89.92% | 88.25% | |||
2019-04-30 | FRCC | 84.67% | 89.53% | 87.03% | |||
2019-04-26 | JDIVA_Textboxes++ | 87.02% | 81.23% | 84.03% | |||
2019-04-29 | Subm190429 | 85.18% | 79.66% | 82.33% | |||
2019-04-28 | gd method | 73.05% | 78.35% | 75.61% | |||
2019-04-28 | CornerNet Multi Scale | 70.35% | 80.19% | 74.95% | |||
2019-04-30 | Textboxes++ detects arbitrary-oriented scene text in a single network forward pass | 60.66% | 90.87% | 72.76% | |||
2019-04-25 | The improved CTPN | 66.83% | 75.87% | 71.07% | |||
2019-04-30 | Scene text detection of polar coordinate regression | 72.54% | 56.44% | 63.48% | |||
2019-04-30 | Multi-scale Pixellink | 50.57% | 32.98% | 39.92% | |||
2019-04-29 | task3 | 7.82% | 8.14% | 7.98% |