VisDoMBench

VisDoMBench is a benchmark dataset that evaluates retrieval-augmented systems on their ability to assemble answers from fragmented evidence in long multimodal documents, including text, tables, and slides.