TextBoxes: A Fast Text Detector with a Single Deep Neural Network

By Minghui Liao, Baoguang Shi, Xiang Bai, Xinggang Wang. 2016

923 citations recorded by the source corpus.

This paper presents an end-to-end trainable fast scene text detector, named TextBoxes, which detects scene text with both high accuracy and efficiency in a single network forward pass, involving no post-process except for a standard non-maximum suppression. TextBoxes outperforms competing methods in terms of text localization accuracy and is much faster, taking only 0.09s per image in a fast implementation. Furthermore, combined with a text recognizer, TextBoxes significantly outperforms state-of-the-art approaches on word spotting and end-to-end text recognition tasks.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Xiang Bai
2. Chunhua Shen
3. Cong Yao
4. Luc Van Gool
5. Dacheng Tao
6. Yu Qiao
7. Xiaogang Wang
8. Andrew Zisserman

In Signal Processing

1. Cong Yao
2. Xiang Bai
3. Minghui Liao
4. Pengyuan Lyu
5. Mingkun Yang
6. Chenglin Liu
7. Chunhua Shen
8. Xin He

Citations in the corpus

• All You Need Is Boundary: Toward Arbitrary-Shaped Text Spotting (2019, methodology): “This assumption is confirmed by the recent end-to-end text spotting methods (Jaderberg et al. 2016; Liao et al. 2017; Liu et al. 2018; Li, Wang, and Shen 2017; He et al. 2018; Busta et al. 2017; Lyu et al. 2018a) that combine the detection and recognition stages with an end-to-end trainable neural n”
• Scene Text Visual Question Answering (2019, methodology): “Several works have been proposed addressing text detection such as [36, 35, 60, 21] which are mostly based on Fully Convolutional Neural Networks.”
• Towards End-to-End Text Spotting with Convolutional Recurrent Neural Networks (2017, methodology): “[17] combine “TextBoxes” and “CRNN” [23], which yield state-of-the-art performance on text spotting task with datasets in [14, 28].”
• TextBoxes++: A Single-Shot Oriented Scene Text Detector (2018, methodology): “Compared with a preliminary study in [13] of this paper, TextBoxes++ replaces the last global average pooling layer with a convolutional layer, which is furthermore beneficial for multi-scale training and testing.”
• Single Shot Text Detector with Regional Attention (2017, methodology): “by the proposed single-shot text detector. Table 2: Exploration study on the ICDAR 2013 dataset, based on the new ICDAR13 standard. Method Recall Precision F-measure SSD [17] 0.60 0.80 0.68 Textboxes [16] 0.74 0.86 0.80 Baseline Model 0.78 0.87 0.82 TAM 0.83 0.87 0.85 HIM 0.82 0.87 0.84 TAM+HIM 0.86”
• Arbitrary-Oriented Scene Text Detection via Rotation Proposals (2017, neutral): “, [6]–[16]).”
• Character Region Awareness for Text Detection (2019, methodology): “To handle this problem, TextBoxes [18] mod-ified convolutional kernels and anchor boxes to effectively capture various text shapes.”
• Deep TextSpotter: An End-to-End Trainable Scene Text Localization and Recognition Framework (2017, background): “Most work however focuses either solely on text localization (detection) [18, 26, 6, 15] or on recognition of manually cropped-out words [7, 24].”
• ABCNet: Real-Time Scene Text Spotting With Adaptive Bezier-Curve Network (2020, methodology): “Method Data Backbone F-measure FPS None Full TextBoxes [22] SynText800k, IC13, IC15, TT ResNet-50-FPN 36.”
• Object Detection in 20 Years: A Survey (2019, methodology): “The first group of methods frame the text detection as a special case of general object detection [251, 348–357].”
• Integrating Scene Text and Visual Appearance for Fine-Grained Image Classification (2017, methodology): “In our case, even though a powerful text reading system TextBoxes [4] is used to spot words in natural images, there are still some errors (see Figure 1 due to large variations in both foreground texts and background objects, and uncontrollable lighting conditions).”
• A Survey of Deep Learning-Based Object Detection (2019, background): “Some typical works [251][252][253] focus on horizontal or nearly horizontal”
• Scene Text Detection and Recognition: The Deep Learning Era (2018, methodology): “or/roi-pooling based methods: (a) Similar to YOLO [117], predicting at each anchor positions. Representative methods include rotating default boxes [91]. (b) Variants of SSD [85], including Textboxes [81], predicting at feature maps of different sizes. (c) Direct regression of bounding boxes [183], ”
• TextSnake: A Flexible Representation for Detecting Text of Arbitrary Shapes (2018, methodology): “Regression based text detection methods [17] mainly draw inspirations from general object detection frameworks.”
• An End-to-End TextSpotter with Explicit Alignment and Attention (2018, background): “Recent approaches cast previous character-based text detection [14, 15, 11, 37] into direct text region estimation [24, 43, 42, 10, 8, 41, 38], which avoids multiple bottom-up post-processing steps by taking word or text-line as a whole detection unite.”
• Real-Time Scene Text Detection With Differentiable Binarization and Adaptive Scale Fusion (2022, methodology): “TextBoxes [25] modified the anchors and the scale of the convolutional kernels based on SSD [30] for text detection.”
• TextNet: Irregular Text Reading from Images with an End-to-End Trainable Network (2018, background): “ine into an end-to-end trainable model. These approaches can be further classied as proposed-based or singleshot models. To tackle multi-scale and orientation problems in text detection, Liao et al. [27] employ SSD framework with specically designed anchor boxes to cover text regions. Liu et al. [31”
• Rotation-Sensitive Regression for Oriented Scene Text Detection (2018, methodology): “Moreover, compared with the most related work [26], RRD also performs better.”