SVTR: Scene Text Recognition with a Single Visual Model

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.

Diffusion in the Dark: A Diffusion Model for Low-Light Text Recognition

“…Liu et al [48] perform text recognition using feature pyramids. We demonstrate that, without any task-specific engineering, we reconstruct fine details to perform robustly in dark, noisy conditions on SOTA text recognition methods [20,19,26,4,58], such as PARSeq [7].…”

Positional Information is a Strong Supervision for Volumetric Medical Image Segmentation

“…AutoSTR [44] searches backbone via neural architecture search (NAS) [45]. More recently, semantic-aware [46], [43], transformer-based [47], linguistics-aware [48], [49], and efficient [50], [51] approaches are proposed to further boost the performance. Although these methods are able to handle irregular, occluded, and incomplete text images, they still have difficulty in recognizing low-resolution images.…”

Morse code detection and recognition algorithm based on YOLO-SVTR

“…The main body of SVTR-LCNet is optimized by SVTR-T 12 , with the first half consisting of the first three stages of PP-LCNet, and the second half consisting of a convolutional pooling layer, two global mixing modules, and a fully connected layer, as shown in Figure 4. Firstly, adjust the cropped Morse code time-frequency map I 1 to 320×16×3, and then input it into the first three stages of PP-LCNet, as shown in Figure 5.…”