The face net model for face recognization used the inception net model. Since the pre-trained weights were used in the assignment I am not able to get a proper intuition regarding the inputs given to the inception network?? is the input provided as a batch of three with a positive, negative, and anchor image??? And which data was the inception net trained on? Were they images of faces??
I don’t know the answers, but they gave you several links in the assignment to references about FaceNet. I’ll bet that the original paper linked in the “References” section at the end of the assignment will cover all that and more.
Notice that they also loaded the model from Keras, so you might find more links if you look up the Keras documention for that model.