Li Xiang ; Wang Wenhai ; Hou Wenbo ; Liu Ruo-Ze ; Lu Tong ; Yang Jian

Abstract

The challenges of shape robust text detection lie in two aspects: 1) mostexisting quadrangular bounding box based detectors are difficult to locatetexts with arbitrary shapes, which are hard to be enclosed perfectly in arectangle; 2) most pixel-wise segmentation-based detectors may not separate thetext instances that are very close to each other. To address these problems, wepropose a novel Progressive Scale Expansion Network (PSENet), designed as asegmentation-based detector with multiple predictions for each text instance.These predictions correspond to different `kernels' produced by shrinking theoriginal text instance into various scales. Consequently, the final detectioncan be conducted through our progressive scale expansion algorithm whichgradually expands the kernels with minimal scales to the text instances withmaximal and complete shapes. Due to the fact that there are large geometricalmargins among these minimal kernels, our method is effective to distinguish theadjacent text instances and is robust to arbitrary shapes. The state-of-the-artresults on ICDAR 2015 and ICDAR 2017 MLT benchmarks further confirm the greateffectiveness of PSENet. Notably, PSENet outperforms the previous best recordby absolute 6.37\% on the curve text dataset SCUT-CTW1500. Code will beavailable in https://github.com/whai362/PSENet.