This dataset was created to benchmark our approach for multi-modal retrieval (images and tables specifically) on pdf documents. It consists of two components:
A set of pdf documents
A csv file which contains questions and the groundtruth document names as well as page numbers
