Shape Robust Text Detection with Progressive Scale Expansion Network

Thanks for your attention. Our method is not good enough to pass the peer review. So we decide to make our method better before we open the source of our code. Sorry about that.

Introduction

Progressive Scale Expansion Network (PSENet) is a text detector which is able to well detect the arbitrary-shape text in natural scene.

Figure 1: Illustration of our overall pipeline.

Figure 2: The procedure of progressive scale expansion algorithm.

Performance

ICDAR 2015

Method	Precision (%)	Recall (%)	F-measure (%)
PSENet-4s	87.98	83.87	85.88
PSENet-2s	89.30	85.22	87.21
PSENet-1s	88.71	85.51	87.08

ICDAR 2017 MLT

Method	Precision (%)	Recall (%)	F-measure (%)
PSENet-4s	75.98	67.56	71.52
PSENet-2s	76.97	68.35	72.40
PSENet-1s	77.01	68.40	72.45

SCUT-CTW1500

Method	Precision (%)	Recall (%)	F-measure (%)
PSENet-4s	80.49	78.13	79.29
PSENet-2s	81.95	79.30	80.60
PSENet-1s	82.50	79.89	81.17

ICPR MTWI 2018 Challenge 2

Method	Precision (%)	Recall (%)	F-measure (%)
PSENet-1s	78.5	72.1	75.2

Results

Figure 3: The results on ICDAR 2015, ICDAR 2017 MLT and SCUT-CTW1500