SCUT-CTW1500 is a dataset that can be used to recognize curved text in scenes. One of the polygon-based curve text detectors, CTD, is able to explore contextual information instead of independently predicting points, making the detection more accurate.

The dataset includes:

• 1500 images (1000 for training and 500 for testing)
• 10,751 cropped images containing text, of which 3,530 images contain curved irregular text
• A large number of images contain text that is horizontal or in different orientations

The dataset is artificially obtained from the Internet, image libraries (such as Google Open-Image), and mobile phone albums.