ROC curves for the networks described in Section 3. Performance of the Zoobot networks are shown in panel (a), Naberrie and MRC networks are shown in (b), NG, 4-layer CNN, and ResNet-18 are shown in (c), and Denselens and LencCLR are shown in (d). The performance of the networks described in Sect. 3.7 are shown in (e), and (f) shows the CBEN networks at two voting values β = 0.3 and β = 0.0001, respectively. The orange diagonal line marked represents the expected performance of an entirely random classifier. The values in the legend of each subplot represent the true positive recovery for each network. Higher performance is represented in general by curves approaching the top-left corner of the plots.