For n-th item, the item on the 1st row, n-th column is the reconstructed item with a normal autoencoder. And the 2nd is reconstructed with a fake autoencoder, which uses the information without the label (digit). And the 3rd is reconstructed with the normal autoencoder, with another label (given 'm' for m-th column).