Efficiently extending single-page models to handle multi-page documents in visual question answering.
