Fig. (6)
Recognition correct rate of different networks and different input.
