Overview - ICDAR2019 Robust Reading Challenge on Arbitrary-Shaped Text

This is a challenge of scene text understanding, which can be broken down into scene text detection, recognition, and spotting problems. The main novelty of this competition resides in the nature of the competition's dataset - the ArT dataset. Specifically, almost a quarter of the text instances in the dataset are arbitrary-shaped as illustrated in Figure 1, which are rarely seen in previous commonly used benchmarks [1, 2, 3]. There are three main tasks in this competition, which will are detailed in the Tasks tab.


Figure 1. Example images of the ArT dataset. Red color binding lines are formed with polygon ground truth format.


ArT is a combination of Total-Text [4], SCUT-CTW1500 [5] and Baidu Curved Scene Text, which were collected with the motive of introducing the arbitrary-shaped text problem to the scene text community. On top of the existing images (3055), more than 7111 images are added to mixture of both datasets, which make ArT one of the larger scale scene text datasets today. There is a total of 10,166 images in the ArT dataset. It is split into a training set with 5603 images, and a testing set of 4563 newly collected images. The ArT dataset was collected with text shape diversity in mind, hence all existing text shapes (i.e. horizontal, multi-oriented, and curved) have high number of existence in the dataset, which makes it an unique dataset since most of the existing datasets [1, 2, 3] were dominated by horizontal and multi-oriented text instances only.

Text instances in the ArT dataset were annotated with (a) quadrilateral bounding boxes, 8, 10 and 12 vertexes polygon bounding box (more details in Tasks tab), and (b) transcription. Both of these annotations cater for the (a) text detection, (b) recognition, and (c) text spotting tasks proposed by this challenge.



  1. Karatzas, Dimosthenis, et al. "ICDAR 2013 robust reading competition."12th IAPR International Conference on Document Analysis and Recognition (ICDAR). IEEE, 2013.
  2. Karatzas, Dimosthenis, et al. "ICDAR 2015 competition on robust reading." 13th IAPR International Conference on Document Analysis and Recognition (ICDAR). IEEE, 2015.
  3. Gomez, Raul, et al. "ICDAR2017 robust reading challenge on COCO-Text." 14th IAPR International Conference on Document Analysis and Recognition (ICDAR). IEEE, 2017.
  4. Ch'ng, Chee Kheng, and Chee Seng Chan. "Total-text: A comprehensive dataset for scene text detection and recognition." 14th IAPR International Conference on Document Analysis and Recognition (ICDAR). Vol. 1. IEEE, 2017.
  5. Yuliang, Liu, Lianwen, Jin, et al. "Curved Scene Text Detection via Transverse and Longitudinal Sequence Connection." Pattern Recognition, 2019.
  6. C. Chng, Y. Liu, Y. Sun, et al, “ICDAR 2019 Robust Reading Challenge on Arbitrary-Shaped Text-RRC-ArT”, in Proc. of ICDAR 2019.

Important Dates

1st January to 1st March

i) Q&A period for the competition,

ii) The launching of initial website

15th Feb to 1st March

i) Competition formal announcement,

ii) Publicity,

iii) Sample training images available,

iv) Evaluation protocol, file formats etc. available.

25th February

i) Evaluation tools ready,

ii) Full website ready.

1st March

i) Competition kicks off officially,

ii) Release of training set images and ground truth.

9th April

Release of the first part of test set images (2271 images),

20th April

i) Release of the second part of test set images (2292 images).

ii) Website opens for results submission

30th April

i) Deadline of the competition and result submission closes(at PDT 23: 59)

ii) Release of the evaluation results.

5th May

i) Submission deadline for 1 page competition report, and the final ranking will be released after results checking.

20th to 25th September

i) Announcement of competition results at ICDAR2019.