What is the importance of the flatten layer?

[abdou31]

Hello, I focus on my project for now, as I see when I read the document that you gave me is that the network is like this:

But when I see the script landmark.py (old version), I noticed that there is an other layer that does not write on the document, I'm talking about the flattening layer. I would like to know is it important to mention that layer or no?                    

[yinguobing]

The flatten layer is required so the input of the following dense layer is a vector other than matrix.

[abdou31]

If I understand what you say , you mean that the input of the first dense layer ( full 1024 ) should be vector. For what reason?

• Is it to input the vector to tensors of the future neural network or just to the second dense layer ( full 2n ) " logits\BiasAdd" ?