Examples of prediction made by the CNN trained with maps at several redshifts z_obs and ζ30. The left panels show the 21 cm δT_b maps. The middle column shows the predicted t_reion fields. The right panels show the difference TRUE-PRED. Hence, the redder it appears on the map, the more the CNN overestimates the true value at the given pixel; instead, the darker it is, the more the CNN underestimates the real value. The true t_reion(r) is shown in Fig. 7.