I am trying to implement a loss function that computes a loss depending on the (unaugmented) data. So far I found an example detailing the process using the model.add_loss() method of a tf.keras.models.Model()

Note

It is important to examine the code as well as the model architecture carefully to ensure the desired functionality is implemented correctly. One of the helpful tools is the plot_model function from tf.keras.utils:

tf.keras.utils.plot_model(
    m,
    to_file='model.png',
    show_shapes=True,
    show_layer_names=True,
)

Output: