Flip Augmentation

Artificially double the number of training images by performing a horizontal flip. If your data is not left-right symmetrical (which will be the case when training humans, subjects), this option should be avoided.