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: AntFin-Cascade Mask R-CNN2023-02-23

Authors: Yangkun Lin, Tao Xu

Affiliation: Ant Group

Description: Our detector is based on Cascade Mask R-CNN. We use ConvNeXt-B as backbone. SynthText800k and
VISD10k are used to pretrain, and then we finetune on ArT, ICDAR2019-MLT and part of LSVT with multi-scale training. Multi-scale testing is used to get the result.

method: I3CL2021-07-05

Authors: Jian Ye, Jing Zhang, Juhua Liu, Bo Du and Dacheng Tao

Affiliation: Wuhan University SigmaLab, JD Explore Academy

Email: leaf-yej@whu.edu.cn

Description: A arbitrary-shaped scene text detector based on Mask R-CNN. In this result, we use ResNeSt-101 as the backbone. Multi-scale training and testing are applied to get the final result. Our training datasets contain SynthText (pretrain), ArT, ICDAR2019-MLT, and part of LSVT.

Ranking Table

Description Paper Source Code
DateMethodRecallPrecisionHmean
2024-05-30CPN (multi-scale)80.78%76.35%78.50%
2023-02-23AntFin-Cascade Mask R-CNN76.02%79.41%77.68%
2021-07-05I3CL73.68%79.35%76.41%
2019-04-30MEGVII_Detection70.56%82.49%76.06%
2019-12-17Tencent TEG OCR73.78%77.85%75.76%
2019-11-04Sogou_OCR71.65%80.27%75.72%
2020-05-21DuXiaoman_OCR71.42%79.04%75.04%
2020-04-22Mask R-CNN70.75%77.84%74.12%
2023-08-09SRFormer (ResNet50-#1seg)67.07%78.53%72.35%
2024-01-18LRANet68.07%76.79%72.17%
2019-04-29ArtDet-v266.63%78.32%72.01%
2022-04-19TextBPN++(ResNet-50 with DCN)68.35%74.94%71.49%
2023-07-08CPNText-DETR(resnet-50)68.17%74.90%71.38%
2019-04-30CUTeOCR65.13%78.79%71.31%
2020-10-01TextFuseNet (ResNeXt-101)65.94%77.40%71.21%
2022-04-21I3CL(ViTAEv2-S)67.86%74.52%71.03%
2022-07-11DPText-DETR (ResNet-50)66.85%75.26%70.81%
2019-05-01NJU-ImagineLab65.04%76.55%70.33%
2023-03-08TD-PPIoU68.41%71.99%70.16%
2023-11-29ESRNet65.48%74.80%69.83%
2019-04-26baseline_polygon65.41%71.59%68.36%
2022-10-31TD-PPIoU (Long-Pretrain)63.93%73.28%68.29%
2022-03-25TextBPN++(ResNet-50)62.78%71.68%66.94%
2021-03-26TextFuseNet (ResNet-50)60.98%72.55%66.27%
2021-04-08AutoCV62.68%69.58%65.95%
2019-04-30DMText_art58.60%75.38%65.94%
2019-04-29SRCB_Art61.15%69.95%65.25%
2019-04-30A scene text detection method based on maskrcnn57.84%74.82%65.24%
2019-04-29Sg_ptd59.15%72.23%65.04%
2019-04-30Fudan-Supremind Detection v361.63%68.21%64.76%
2019-04-28CLTDR58.14%72.83%64.66%
2019-04-28 Alibaba-PAI62.00%67.01%64.41%
2023-05-15dp_pq_nn57.53%71.71%63.84%
2023-05-15dp_nn58.23%68.29%62.86%
2019-04-24fdu_ai53.48%71.27%61.11%
2019-04-30CCISTD53.40%71.37%61.09%
2021-04-28NN_Chinese_and_euro654.78%68.16%60.74%
2019-05-01TextMask_V162.09%59.23%60.63%
2019-04-30MaskRCNN_Text56.10%65.92%60.61%
2019-04-30TEXT_SNIPER57.52%61.32%59.36%
2019-04-30MaskDet55.43%63.22%59.07%
2019-04-30Mask RCNN62.71%55.82%59.07%
2019-04-16Art_test_baseline_task153.25%61.04%56.88%
2019-04-30TMIS45.80%73.81%56.53%
2021-04-23HOCRA54.30%58.86%56.49%
2019-04-29CRAFT53.14%59.55%56.16%
2019-04-22MFTD: Mask Filters for Text Detection52.41%59.92%55.92%
2019-04-30QAQ48.86%64.49%55.60%
2019-04-25Art detect by vivo47.44%67.00%55.55%
2019-04-29PAT-S.Y48.46%61.53%54.22%
2021-04-28NN_euro643.21%71.39%53.84%
2019-04-30DMCA50.33%54.32%52.25%
2019-04-26TP42.22%63.94%50.86%
2019-04-22mask rcnn43.67%58.76%50.11%
2019-04-28Improved Progressive scale expansion Net41.79%60.70%49.50%
2019-05-01Unicamp-SRBR-PN-142.90%50.67%46.46%
2019-04-27TextCohesion_134.80%54.26%42.40%
2021-04-29HOCRA_base29.36%72.45%41.78%
2019-04-23142.27%41.08%41.66%
2019-04-26RAST: Robust Arbitrary Shape Text Detector27.36%54.86%36.51%
2019-04-30EM-DATA27.96%38.01%32.22%
2021-03-31inception baseline16.97%33.66%22.57%
2019-04-30MSR0.06%0.08%0.07%

Ranking Graphic