Next, we set up a function to load our data and subsequently perform needed transforms. This functions uses a few tools that exist within torchvision.

Dataset

torchvision.datasets.ImageFolder is a generic data loader where images are structured as follows:

Note the directories C0, C2, and C4 are the names of the three classes in our classifier and, in this case, refer to low, medium and high levels of damage. Another noteworthy feature of torchvision's datasets is that images are not immediately loaded into memory. They are loaded one by one as needed.

Transforms

torchvision.transforms.Compose applies a series of transformations to your data. In our function, we use Compose to perform a series of two transforms:

• torchvision.transforms.Resize which resizes your image to the specified dimension
• torchvision.transofrms.ToTensor converts you images from PIL or numpy arrrays to a torch tensor.

In the function below, we use ImageFolder to load our train, validation and testing data and perform the needed transformations as described above.