CNN score (S_CNN) distribution for the different datasets. Top panel: S_CNN distribution of both test sets: lens simulations (dashed line) and real confirmed and candidate lenses (solid line). Both datasets contain images labeled as lenses (purple lines) and real LRGs labeled as non-lenses (red lines). The two test-sets are normalized to their corresponding maximum value in the distribution. We shifted the x-axis of the second test set distribution by 0.02 for clarity. Bottom panel: S_CNN distribution for the objects in the parent sample with scores above 0.5. We crop the x-axis for visualization as 99% of the sample is below 0.5.