Large image datasets: A pyrrhic win for computer vision? - 42Papers