Where does a computer vision model make mistakes? Using interactive visualizations to find where and how CV models can improve