Fig 4. Data augmentation.

Examining the accuracy of the CNN as a function of a) varying the training data size by adding images from biologically distinct datasets (New datasets) or by adding images randomly from the full set of images (Train/test split), and b) transformation of the data by image rotations and reflections. In (a), the two empty circles represent the inclusion of the datasets from empty (germ-free) zebrafish intestines.