Overview - ICDAR 2019 Robust Reading Challenge on Reading Chinese Text on Signboard

Texts in natural images carry much important semantic information. Automatically detecting and recognizing text in natural scene images has been receiving increasing attention due to its wide range of applications. It’s also an important prerequisite for many content-based image analysis tasks.

In recent years, more and more datasets including Chinese have been proposed for natural scene text reading tasks, such as MSRA-500[1], RCTW[2], SCUT-CTW1500[3], CTW[4], etc. The layout and arrangement of Chinese characters in signboards are obviously different from that in other scenes. For the sake of aesthetics appearance or highlighting certain elements, the Chinese characters can be arranged in various layouts with various fonts. For instance, characters in one word or text line can be in diverse fonts, various orientations or distributed in irregular layouts (see Figure 1a, 1b 1c). Previous word-based reading systems are hard to deal with irregular layouts such as triangles, as shown in Figure 2. In these cases, characters and their relations should be considered to read them semantically. In our dataset, we annotate not only the text lines, but also the characters, including positions and character codes, which can inspire and power new algorithms that make use of the arrangement of characters.

ReCTS_f21.png

Figure 1. Text images with various layouts, fonts, and orientations.

 

ReCTS_f1.png

Figure 2. An example of reading text of irregular layout.

We collect and construct a practical and challenging multi-orientation natural scene text dataset (ReCTS) with 25,000 images, which consist of lots of signboards. In the dataset, all text lines and characters are labeled with locations and character codes. Besides, four tasks are presents for this competition: (1) end-to-end recognition on the signboards, (2) text line localization on the signboards, (3) character recognition on the signboards, (4) text line recognition on the signboards.

We hope that the dataset and tasks could greatly promote the research in Chinese text detection and recognition.

Awards

The prize for ICDAR 2019-ReCTS is $15,000 in total, sponsored by Meituan.

Note: prizes are only for top 3 student contestants.

Task 1. Character recognition on the signboard,  $1,500/$1,000/$600 for top 3 winners.

Task 2. Text line recognition on the signboards,  $1,500/$1,000/$600 for top 3 winners.

Task 3. Text line localization on the signboards,  $1,500/$1,000/$800 for top 3 winners.

Task 4. End-to-end recognition on the signboards,   $3,000/$1,500/$1,000 for top 3 winners.

Reference

[1] MSRA-500:C. Yao, X. Bai, W. Liu, Y. Ma, Z. Tu. Detecting Texts of Arbitrary Orientations in Natural Images. CVPR, 2012.  

[2] RCTW:Shi B , Yao C , Liao M , et al. ICDAR2017 Competition on Reading Chinese Text in the Wild (RCTW-17)[J]. 2017.

[3] SCUT-CTW1500:Yuliang L , Lianwen J , Shuaitao Z , et al. Detecting Curve Text in the Wild: New Dataset and New Solution[J]. 2017.

[4] CTW:Yuan T L , Zhu Z , Xu K , et al. Chinese Text in the Wild[J]. 2018.

Important Dates

1 March 2019: Web site ready and Registration open

18 March 2019: Training set available

12 April 2019: The first part of test set available

20 April 2019: The second part of test set available, and Website opens for result submission

30 April 2019: Submission of results deadline

10 May 2019: Deadline for providing short descriptions of the participating methods

20-25 September 2019: Results presentation