Synthetic dataset of folded and projectively distorted documents

Description The dataset contains documents with three types of distortions: 2fold (95 images) cross_fold (100 images) projectively distorted (325 images) Each type of distortion has its own folder. All images was created using scans from DocUnet benchmark. Dataset Structure Folder "dataset" contains sorted in three folders (concerning its distortion type) document images. Each image has corresponding JSON file with document boundaries. Folders 2fold and cross_fold contains cropped images with margin 4, but proj_dist contains images without cropping.