As mentioned above, PyTorch's dataset features load images one by one. When training deep learning models, we typically use a batch gradient descent algorithm to optimize our network and thus we will need to load random images in our specified batch sizes at each gradient descent step. PyTorch's DataLoader is an iterable that automatically performs and loads the data need at each iteration via a simple API. In the function, we instantiate the DataLoader for our train, test and validation datasets.