Data augmentation

Technique called “data augmentation” refers to generation of artificial data in order to increase the training size so that learning model improves its prediction accuracy. The folder named “original_data” contains dog and cat images from “cifar10” dataset. There is also a python script call “generate_augmented_data.py”, that creates two additional images for each original image by rotating the original image randomly with a degree between -45 degrees and 45 degrees. The script will create a folder “augmented_data/train/” that will contain the original images + the augmented data. We have compared prediction accuracy for the model trained on original data and augmented data. Three different learning rates of optimizer were used: 0.001, 0.005, 0.05.