KVQA Question Answering Dataset

KVQA, short for Knowledge-aware VQA, is a dataset about question answering. The dataset contains 183,000 manually verified question-answer pairs involving more than 18,000 people and 24,000 images. The questions in the dataset require multi-entity, multi-relation, and multi-hop reasoning on KG to find answers.

To enable visual named entity linking, it also provides a support set containing 69,000 reference images of people collected from Wikidata.