pan_pp

PAN++: Towards Efficient and Accurate End-to-End Spotting of Arbitrarily-Shaped Text

Introduction

@article{wang2021pan++,
  title={PAN++: Towards Efficient and Accurate End-to-End Spotting of Arbitrarily-Shaped Text},
  author={Wang, Wenhai and Xie, Enze and Li, Xiang and Liu, Xuebo and Liang, Ding and Yang, Zhibo and Lu, Tong and Shen, Chunhua},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2021},
  publisher={IEEE}
}

Results and Models

Text Detection

• ICDAR 2015

Method	Backbone	Finetune	Precision (%)	Recall (%)	F-measure (%)	Config	Download
PAN++ detection only	ResNet18	N	84.4	78.5	81.3	config	model

End-to-End Text Spotting

• ICDAR 2015

Method	Backbone	Finetune	Vocabulary	Precision (%)	Recall (%)	F-measure (%)	Config	Download
PAN++ 736 joint train	ResNet18	N	N	83.6	54.0	65.6	config	model
PAN++ 736 joint train	ResNet18	N	G	82.3	55.8	66.5	config	model
PAN++ 736 joint train	ResNet18	N	W	90.1	63.9	74.8	config	model
PAN++ 736 joint train	ResNet18	N	S	92.2	70.3	79.8	config	model
_
PAN++ 896 joint train	ResNet18	N	N	82.0	56.0	66.6	config	model
PAN++ 896 joint train	ResNet18	N	G	81.5	57.6	67.5	config	model
PAN++ 896 joint train	ResNet18	N	W	90.1	63.9	74.8	config	model
PAN++ 896 joint train	ResNet18	N	S	92.2	70.3	79.8	config	model

Todo:

• Models and configs on other datasets.