Trained on dataset of 20 images with different aspect ratio, angles and face expressions