Dataset statistics. # imgs: number of unique images; sent. L: average sentence length; # W, # WP: resp., number of words and word pairs.