1 paper · 1 filter
Rongjian Gu, Wengang Zhou, Junyu Xiong +4
Multi-page document visual question answering requires locating sparse evidence at both the page and region levels. Existing approaches typically emphasize one level over the other…