Combining 2 networks :Testing a network and using its outputs for training another network

[p.Paul]

I have RGB images as input  for network-1
I have corresponding input2=(depth image + features of network 1) as input to the network-2

Why I see a problem  to combine them is because I don't have ground truths for network -1 task, I have only GT for network2
Is it possible to define network-1 only for testing, so that there are no labels involved,  and network-2 for train and test and write it all in a single prototxt file?

I want to know if Testing a network(already well trained) and using its outputs for training another network is possible.
Could you please share some ideas or thoughts on this? It would be really helpful.
 
I don't know if it has something to do with multi-task learning, because, two tasks are on two different networks.
I don't know if share layers will help with this, I could not find a detailed explanation on this  regarding my task. Any suggestion would be appreciated

Thank you very much in advance

[p.Paul]

Hello guys, any thoughts would be helpful