Semantic-Information Space Sharing Interaction Network for Arbitrary Shape Text Detection

Arbitrary shape text detection is a challenging task due to significant variations in text shapes, sizes, and aspect ratios. Previous approaches relying on single-level feature map generated through a top-down fusion of different feature levels have limitations in harnessing high-level semantic information and expressing multi-scale features. To address these challenges, this paper introduces a novel arbitrary shape scene text detector called the Semantic-information Space Sharing Interaction Network (SSINet). The proposed network leverages the Semantic-information Space Sharing Module (SSM) to generate a single-level feature map capable of expressing multi-scale features with rich semantic and prominent foreground, enabling effective processing of text-related information. Experimental evaluations on three benchmark datasets, namely CTW-1500, MSRA-TD500, and ICDAR2017-MLT, validate the effectiveness of our method. The proposed SSINet achieves impressive results with an F-score of 86.0% on CTW-1500, 89.1% on MSRA-TD500, and 72.4% on ICDAR2017-MLT. The code will be available at https://github.com/123cjjjj/SSINet .