True vs predicted maps obtained from the full dataset in the test set, for z_obs = 11 and for both models. The red lines stand for the perfect correlation. The bottom and left histograms are the mean and the standard deviation of the residual r: r = Predicted−True, in both vertical and horizontal directions. The bottom histogram stands for the learning error, while the side histogram stands for the recovery uncertainty. In practice, the recovery uncertainty is the only accessible estimator since the ground truth will not be accessible.