In order to facilitate a new document recognition research, we introduce a Distorted Document Images dataset (DDI-100).
To create the dataset we collected 6658 unique document pages, and extended it by applying different types of distortions and geometric transformations.
In total, DDI-100 contains 99870 document images together with text masks, stamp masks, text and character locations in terms of bounding boxes.
Direct link and detailed description is in dataset directory
In scripts directory you can find some python scripts with usefull functions and classes for working with dataset.
Preprint can be found here: https://arxiv.org/abs/1912.11658