psenet

Shape Robust Text Detection with Progressive Scale Expansion Network

Introduction

@inproceedings{wang2019shape,
  title={Shape Robust Text Detection with Progressive Scale Expansion Network},
  author={Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={9336--9345},
  year={2019}
}

Results and Models

• Total-Text

Method	Backbone	Finetune	Precision (%)	Recall (%)	F-measure (%)	Config	Download
PSENet	ResNet50	N	87.3	77.9	82.3	config	model
PSENet (paper)	ResNet50	N	81.8	75.1	78.3	-	-
PSENet	ResNet50	Y	89.3	79.6	84.2	config	model
PSENet (paper)	ResNet50	Y	84.0	78.0	80.9	-	-

• CTW1500

Method	Backbone	Finetune	Precision (%)	Recall (%)	F-measure (%)	Config	Download
PSENet	ResNet50	N	82.6	76.4	79.4	config	model
PSENet (paper)	ResNet50	N	80.6	75.6	78.0	-	-
PSENet	ResNet50	Y	84.3	78.9	81.5	config	model
PSENet (paper)	ResNet50	Y	84.8	79.7	82.2	-	-

• ICDAR 2015

Method	Backbone	Finetune	Scale	Precision (%)	Recall (%)	F-measure (%)	Config	Download
PSENet	ResNet50	N	S: 736	83.6	74.0	78.5	config	model
PSENet	ResNet50	N	S: 1024	84.4	76.3	80.2	config	model
PSENet (paper)	ResNet50	N	L: 2240	81.5	79.7	80.6	-	-
PSENet	ResNet50	Y	S: 736	85.3	76.8	80.9	config	model
PSENet	ResNet50	Y	S: 1024	86.2	79.4	82.7	config	model
PSENet (paper)	ResNet50	Y	L: 2240	86.9	84.5	85.7	-	-

• SynthText

Method	Backbone	Config	Download
PSENet	ResNet50	config	model