UsaTax5-dataset

Dataset summary for UsaTax5 The UsaTax5 dataset includes 186 images in total of USA forms and their annotations. There are two folders in the root directory of the dataset: images and markup. Each of them contains subfolders with files divided into categories based on USA form's type (8919, 8839, 8865-H, 8915-F, 1040) and their pages. Each image of the dataset contains a single USA form. Distribution of images among subsets of the UsaTax5 dataset: Type Count of elements Count of pages 8919 15 1 8839 30 2 8865-H 15 1 8915-F 84 4 1040 42 2 Annotation format The main ground truth file (see f8839-1-1.png.json as example) provides the following information about a single USA form within the image: key document_type: the type of presented tax form; key template_quad: location of the document’s quadrangle; key props: collection of properties, containing information regarding the document items within the associated keys. The supplementary ground truth file (see f8839-1-1.png_quads.json as example) provides the information about properties quadrangles, listed in props_quads JSON node. The vertices of all the quadrangles are numbered clockwise from the top left corner. UsaTax5-forms This folder contains descriptions of forms, written in simplified formal language, suitable for their efficient automatic processing.