*3.2. Data Augmentation*

To eliminate overfitting, data augmentation was performed using the ImageDataGenerator class available in the Keras package. The rotation and flip operations were applied to the datasets. The data prepared in this way were used to check whether data augmentation improved the results that were obtained on the validation dataset. Example images that were obtained as a result of data augmentation are presented in Figure 6. The results of network training based on augmented data are presented in Section 4.

**Figure 6.** Example of augmented data: (**a**) images (**b**) masks.
