ABSTRACT
This chapter addresses model training. It follows the following steps to get model trained and ready, in order to perform gap filling: First, download the code of the implementation from the author’s repository. Second use the code to train the model. Model weights will be saved in checkpoint files. Third export a trained Saved Model from an existing checkpoint. Fourth use the Saved Model with the OTBTF Tensor flow Model Serve application to perform the optical image restoration. Many practitioners like to have full control of their training process, and will prefer the TensorFlow Python API rather than using the Tensor flow Model Serve application. Indeed, Tensor flow Model Serve hides the complexity of the training and provides the user an easy interface to train models on their geospatial data, but with a drastically simplified training approach.
