SVTR: Scene Text Recognition with a Single Visual Model

TL;DR

SVTR introduces a simplified scene text recognition model that eliminates the need for sequence modeling by using a hierarchical patch-wise image tokenization approach, achieving high accuracy and efficiency on English and Chinese datasets.

Contribution

The paper presents a novel single visual model, SVTR, that replaces traditional hybrid architectures with a patch-wise tokenization framework for scene text recognition.

Findings

SVTR-L achieves competitive accuracy on English text recognition.

SVTR outperforms existing methods in Chinese text recognition.

SVTR-T offers a smaller, faster model with good accuracy.

Abstract

Dominant scene text recognition models commonly contain two building blocks, a visual model for feature extraction and a sequence model for text transcription. This hybrid architecture, although accurate, is complex and less efficient. In this study, we propose a Single Visual model for Scene Text recognition within the patch-wise image tokenization framework, which dispenses with the sequential modeling entirely. The method, termed SVTR, firstly decomposes an image text into small patches named character components. Afterward, hierarchical stages are recurrently carried out by component-level mixing, merging and/or combining. Global and local mixing blocks are devised to perceive the inter-character and intra-character patterns, leading to a multi-grained character component perception. Thus, characters are recognized by a simple linear prediction. Experimental results on both English and Chinese scene text recognition tasks demonstrate the effectiveness of SVTR. SVTR-L (Large) achieves highly competitive accuracy in English and outperforms existing methods by a large margin in Chinese, while running faster. In addition, SVTR-T (Tiny) is an effective and much smaller model, which shows appealing speed at inference. The code is publicly available at https://github.com/PaddlePaddle/PaddleOCR.