WildReceipt is a collection of receipts. It contains, for each photo, a list of OCRs - with the bounding box, text, and class. It contains 1765 photos, with 25 classes, and 50000 text boxes. The goal is to benchmark "key information extraction" - extracting key information from documents
https://arxiv.org/abs/2103.14470
