how to build a condition network ??

[Johnson-yue]

Hi , I want to know how to build a condition network ?

My idea is :

       Condition is a label ,so I want to use one-hot input and fc_layer in condition network.

      Main network is full convolution network with 4 HaarDownsampling.  

This is failed ,but condtion and featuredon't agree : [(512,) ] vs [(1, 64, 64)]
I think the output of condition network must be same shape as main network feature?

I can understand colorization_minial_example experiments code it is the same as paper said, but colorization_cINN, have many new functions ,so Can you write a blog about it?