Bridging the Gap Between End-to-End and Two-Step Text Spotting

• DOI: 10.48550/arxiv.2404.04624
• Publication Date: 2024-04-06
• Authors: Mingxin Huang; Hongliang Li; Yuliang Liu; Xiang Bai; Lianwen Jin
• Keywords: Spotting; End-to-end principle

Modularity plays a crucial role in the development and maintenance of complex systems. While end-to-end text spotting efficiently mitigates issues error accumulation sub-optimal performance seen traditional two-step methodologies, methods continue to be favored many competitions practical settings due their superior modularity. In this paper, we introduce Bridging Text Spotting, novel approach that resolves suboptimal while retaining To achieve this, adopt well-trained detector recognizer are developed trained independently then lock parameters preserve already acquired capabilities. Subsequently, Bridge connects locked through zero-initialized neural network. This network, initialized with weights set zeros, ensures seamless integration large receptive field features detection into recognizer. Furthermore, since fixed cannot naturally acquire optimization features, Adapter facilitate efficient learning these features. We demonstrate effectiveness proposed method extensive experiments: Connecting latest achieved an accuracy 83.3% on Total-Text, 69.8% CTW1500, 89.5% ICDAR 2015. The code is available at https://github.com/mxin262/Bridging-Text-Spotting.