Data preprocessing

Data preprocessing is an essential step for a deep learning pipeline. The HAR and FER2013 datasets are preprocessed well. However, the downloaded image files for the second dataset of use case two are not preprocessed. As shown in the preceding image, the images are not uniform in size or pixels and the dataset is not large in size; hence, they require data augmentation. Popular augmentation techniques are flip, rotation, scale, crop, translation, and Gaussian noise. Many tools are available for each of these activities. You can use the tools or write their own script to do the data augmentation. A useful tool is Augmentor, a Python library for machine learning. We can install the tool in our Python and use it for augmentation. The following code (data_augmentation.py) is a simple data augmentation process that executes flipping, rotation, cropping, and resizing...