Performance of the final network on the test set T. Left hand side: histograms of the reference and predicted redshift distributions in red and blue, respectively. Right hand side: a 1:1 comparison of reference and predicted redshifts. The red line shows the median predicted redshift per bin and the gray bands the 1σ and 2σ confidence levels. While the red line follows the black dashed reference line for low redshift very nicely, NetZ_main tends to underpredict the high end.