GroundCap: A Visually Grounded Image Captioning Dataset