Sharing layers with multiple branches

Question

Is it possible to share more complicated module with branching, let's say inception module for example?

I found answer for sharing simple sequential module here but it doesn't seem to be sufficient for my purposes.

I need to be able to do something like this:

def shared_module():
    a = conv()
    a = conv(a) 
    b = conv()
    p = concat([a, b])

shared = shared_module()
    m_x = shared(Input(shape))
    m_y = shared(Input(shape))

Answer 1

Yes, you can do any type of branching with Keras.

You should check the Functional api. It's exactly what you need. Also in the keras examples you'll find practical uses of it.