Curved scene text detection via transverse and longitudinal sequence connection

By Yuliang Liu, Lianwen Jin, Shuaitao Zhang, Canjie Luo, Sheng Zhang. Pattern Recognition, 2019

307 citations recorded by the source corpus.

Abstract Curved text detection is a difficult problem that has not been addressed sufficiently. To highlight the difficulties in reading curved text in a real environment, we constructed a curved text dataset called CTW1500, which includes over 10,000 text annotations in 1500 images, and used it to formulate a polygon-based curved text detector that can detect curved text without using an empirical combination. With the seamless integration of recurrent transverse and longitudinal offset connection, our method explores context information instead of predicting points independently, resulting in smoother and more accurate detection. Our approach is designed as a universal method, meaning it can be trained using rectangular or quadrilateral bounding boxes, requiring no extra effort. Experimental results on the CTW1500 dataset and Total-text demonstrated that our method with only a light backbone can outperform state-of-the-art methods by a large margin. Our method also achieved state-of-the-art performance on the MSRA-TD500 dataset, demonstrating its promising generalization ability. Code, datasets, and label-tool are available at https://github.com/Yuliang-Liu/Curve-Text-Detector .

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/a440bcf181c7f7b072ef178a067497c55805685a

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

In Media Technology

1. Cong Yao
2. Xiang Bai
3. Minghui Liao
4. Zhaoyi Wan
5. Lianwen Jin
6. Kai Chen
7. Yuliang Liu
8. Chunhua Shen

Citations in the corpus

• LVLM-EHub: A Comprehensive Evaluation Benchmark for Large Vision-Language Models (2023, unknown): “(cid:2) LVLM pre-trained with extensive, high-quality data can achieve competitive performance in multimodal comprehension: Lynx, which is pre-trained with 124 M massive high-quality data covering various multimodal tasks achieves competitive performance in quantitative evaluation and the best…”
• Structured Multimodal Attentions for TextVQA (2020, methodology): “The SBD model is pretrained on a 60k dataset, which consists of 30, 000 images from LSVT [40] training set, 10, 000 images from MLT 2019 [31] training set, 5, 603 images from ArT [10] which contains all the images of SCUT-CTW1500 [27] and Total-text [9,11], and the rest of 14, 859 images are selecte”
• Arbitrary Shape Scene Text Detection With Adaptive Text Region Representation (2019, methodology): “Five benchmarks are used in this paper for performance evaluation, which are introduced in the following: • CTW1500 : The CTW1500 dataset [17] • MSRA-TD500 : The MSRA-TD500 dataset [29] contains 200 test images and 300 training images, that contain arbitrarily-oriented texts in both Chinese and…”
• Real-time Scene Text Detection with Differentiable Binarization (2019, background): “Given a text image, each polygon of its text regions is described by a set of segments: n is the number of vertexes, which may be different in different datasets, e.g, 4 for the ICDAR 2015 dataset (Karatzas et al. 2015) and 16 for the CTW1500 dataset (Liu et al. 2019a).”
• Real-Time Scene Text Detection With Differentiable Binarization and Adaptive Scale Fusion (2022, methodology): “CTW1500 dataset [32] mainly focuses on curved text.”
• ImageBind-LLM: Multi-modality Instruction Tuning (2023, methodology): “We evaluate ImageBind-LLM on 12 representative OCR datasets, including IIIT5K [61], ICDAR 2013(IC13) [62], ICDAR 2015 (IC15) [63], Total-Text [64], CUTE80 [65], Street View Text (SVT) [66], SVTP-Perspective (SVTP) [67], COCO-Text [68], WordArt [69], SCUT-CTW1500 (CTW) [70], Heavily Occluded Scene Te”
• On the General Value of Evidence, and Bilingual Scene-Text Visual Question Answering (2020, methodology): “Specifically, images annotated with English questions and answers are obtained from Total-Text [5], ICDAR 2013 [14], ICDAR 2015 [13], CTW1500 [20], MLT [25], and COCO Text [32].”
• MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models (2023, unknown): “The images are from [34] and the instruction-answer pairs are manually designed.”
• ABCNet: Real-Time Scene Text Spotting With Adaptive Bezier-Curve Network (2020, methodology): “Thus, our motivation is to design a simple yet effective end-to-end framework for spotting oriented or curved scene text in images [4, 27], which ensures fast inference time while achieving an on par or even better performance compared with state-of-the-art methods.”