Fig. 4

The 11 ways of data augmentation that were compared. The flowchart hierarchically illustrates the steps to implement the final 11 ways to apply data augmentation. Colored packets represent and are proportional to parts of the dataset. Red, blue, and orange packets represent independent training, validation, and testing data, respectively. Purple packets represent training and validation data when some training images are derived by augmenting some parent validation images and vice versa. Brown packets represent the three subsets when each subset contains some augmentation derivatives of some parent images in the other two subsets. Dashed-outline box = starting point; dotted-outline boxes = intermediate steps; solid-outline boxes = final 11 ways to apply data augmentation which were evaluated