* This dataset supports online use.Click here to jump.

The SynthText dataset consists of natural scene images containing words. It is mainly used for text detection in natural scenes. The dataset consists of 800,000 images and has approximately 8 million synthetic word instances.

Each text instance is annotated with text string, word-level, and character-level bounding boxes.

The SynthText dataset was published by Gupta, A., Vedaldi, A. and Zisserman, A. of the Visual Geometry Group, Department of Engineering Science, University of Oxford, at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) in 2016.