The relevant table on the results page is Table 3, which is detection performance. Classification is actually an easier problem (see Table 1), in part because t

The relevant table on the results page is Table 3, which is detection performance. Classification is actually an easier problem (see Table 1), in part because the types of scenes in which different classes appear are often quite different, making it easy to avoid some "easy" mistakes.