The workflow reads the downloaded image files and partitions into a training and hold-out set. Images are then split into multiple 64x64px patches that can be used by the VGG16 network in the following model.