Fig. 7

Model validation accuracies. Validation accuracies for the 24 studied models. The four different symbol shapes/colors correspond to the four networks. Error bars are binomial exact 95% confidence intervals. Horizontal axis labels stand for different ways of applying augmentation: A = Make three sets, then augment validation; B = Make three sets; C = Make three sets, then augment training; D = Make three sets, then augment both training and validation; E = Separate test set, augment the rest, then make two other sets; F = Augment all before making three sets. CI = confidence interval