method: NSTD-iFLYTEK2019-10-10

Authors: iFLYTEK

Affiliation: iFLYTEK

Description: Natural scene text detector(NSTD-iFLYTEK) is based on MaskRcnn with resnet-101. Only ICDAR2019 datasets are used for training, including Rects, LSVT, MLT and Art. Multi-scale training and single-scale testing are used to generate the final result, no model ensemble.
name and organization:
Jian Dong(董健) iFLYTEK(科大讯飞)
Fengren Wang(王烽人) iFLYTEK(科大讯飞)
Jiajia Wu(吴嘉嘉) iFLYTEK(科大讯飞)
Yin Lin(林垠) iFLYTEK(科大讯飞)
Lou Shun(娄舜) iFLYTEK(科大讯飞)
Jinshui Hu(胡金水) iFLYTEK(科大讯飞)

method: SANHL_v42019-05-01

Authors: Yuliang Liu, Hao Chen, Shuaitao Zhang, Lele Xie, Dezhi Peng, Weihong Ma, Tong He, Chongyu Liu, Xiangle Chen, Canjie Luo, Qingxiang Lin, Lianwen Jin, Chunhua Shen, Yaqiang Wu, Liangwei Wang

Description: This method utilizes a sequential-free box dicretization method to localize the text instances. Only public datasets are used for pre-training, including LSVT, ArT and MLT. Multi-scale testing and model ensemble are used to generate the final result. The result is submitted by the researchers from South China University of Technology, Northwestern Polytechnical University, The University of Adelaide, Lenovo and Huawei. The researchers are Yuliang Liu, Hao Chen, Shuaitao Zhang, Lele Xie, Dezhi Peng, Weihong Ma, Tong He, Chongyu Liu, Xiangle Chen, Canjie Luo, Qingxiang Lin, Lianwen Jin, Chunhua Shen, Yaqiang Wu and Liangwei Wang.

单位:华南理工大学,阿德莱德大学,西北工业大学,联想,华为。
作者:刘禹良,陈昊,张帅涛,谢乐乐,彭德智,马伟洪,贺通,刘崇宇,陈向乐,罗灿杰,林庆祥,金连文,沈春华,武亚强,王靓伟。

method: Tencent-DPPR Team2019-05-01

Authors: Longhuang Wu, Shangxuan Tian, Chang Liu, Wenjie Cai, Jiachen Li, Chunchao Guo, Sicong Liu, Haoxi Li, Hongfa Wang, Hongkai Chen, Qinglin lu, Xucheng Yin, Lei Xiao

Description: We are Tencent-DPPR (Data Platform Precision Recommendation) team. In detection stage, we use LSVT dataset to pretrain our model and provided ReCTS dataset to train the text detector. During training, we use multi-scale training policy.
Our text detector is based on two-stage method. In backbone part, we use ResNet101 as feature extractor. In FPN part, we designed a policy to help proposals select feature pyramid layers to extract features instead of choosing one layer according to box sizes.
In detection ensemble part, we apply a multi-scale test method with different backones. When ensembling all the results, we develop an approach to vote boxes after scoring each box.

Ranking Table

Description Paper Source Code
DateMethodRecallPrecisionHmean
2019-10-10NSTD-iFLYTEK93.17%93.62%93.40%
2019-05-01SANHL_v493.97%92.76%93.36%
2019-05-01Tencent-DPPR Team93.46%92.59%93.03%
2019-04-29Amap-CVLab93.41%91.62%92.50%
2019-04-30HUST_VLRGROUP93.51%89.15%91.27%
2019-04-30maskrcnn_text91.96%90.09%91.02%
2019-04-30Task3-re590.03%91.65%90.83%
2019-04-22oo91.56%90.08%90.81%
2019-04-23A region proposal and fcn model based method 88.64%92.72%90.64%
2019-04-30Mask R-CNN89.84%91.41%90.62%
2019-04-30COLD AND COOL90.99%89.59%90.28%
2019-04-26baseline_0.793.66%86.35%89.86%
2019-04-30pursuer86.13%92.72%89.31%
2019-04-29CLTDR88.92%88.70%88.81%
2020-05-18DST86.63%89.92%88.25%
2019-04-30CRAFT85.33%89.38%87.31%
2019-04-30FRCC84.67%89.53%87.03%
2019-04-25EAST检测网络82.27%88.49%85.27%
2019-04-26JDIVA_Textboxes++87.02%81.23%84.03%
2019-04-30FFLOVE88.52%79.32%83.66%
2019-04-29Subm19042985.18%79.66%82.33%
2019-04-23PSENet_v183.16%80.77%81.94%
2019-04-30Sogou_MM96.17%69.20%80.48%
2019-04-30WHUT79.53%79.36%79.45%
2019-04-30PixelBased Prediction86.02%70.68%77.60%
2019-10-31Cluster75.80%77.05%76.42%
2019-04-28gd method73.05%78.35%75.61%
2019-04-28CornerNet Multi Scale70.35%80.19%74.95%
2019-04-30Textboxes++ detects arbitrary-oriented scene text in a single network forward pass60.66%90.87%72.76%
2019-04-25The improved CTPN66.83%75.87%71.07%
2019-04-30Scene text detection of polar coordinate regression72.54%56.44%63.48%
2019-04-30Multi-scale Pixellink50.57%32.98%39.92%
2019-04-29task37.82%8.14%7.98%

Ranking Graphic