Prediction of the Bayesian graph-convolutional neural network (GCNN) for the Fermi photon-count map. The shaded regions show the predictive (aleatoric and epistemic summed in quadrature) 1σ uncertainty. The markers with error bars (68% credible intervals) indicate the Non-Poissonian template fit (NPTFit) estimates for comparison. The NN predictions for the GCE flux in the Fermi map are similar in magnitude to those of NPTFit, but the GCE is almost entirely attributed to the smooth dark matter template. To view the plot, open it with a browser such as Google Chrome or Firefox. Templates can be (de-)activated by clicking on the colored rectangles. For zooming and panning, click on the buttons in the lower left corner.