Towards recognizing "cool": can end users help computer vision recognize subjective attributes of objects in images?