In this notebook, we will build and train our own U-Net image segmentation architecture, in order to segment satellite images.

We will use the DroneDeploy Segmentation Dataset from https://github.com/dronedeploy/dd-ml-segmentation-benchmark of aerial photographs taken by drones, and segmentation labels for each scene. See that link for more information on the dataset.

We have already made an EDA on the dataset and we will just focuse here on the model.