I am trying to fuse two CNN through dense layers, each dense layer has variable size. So I want to use another global dense layer to fuse individual CNN dense layers. Is it correct, How can I implement this, is concatenating necessary or we can directly send both dense layer output to global dense layer ?
Could you write down some pseudo-code or upload a small sketch showing how these layers should be concatenated?
This would help a bit understanding your use case.
PS: Tagging certain people might demotivate others to post in your topic, so I’m not a huge fan of it.
Based on the sketch it looks like the FC-1-4 layers should all be concatenated and passed to the FC-Softmax layer.
Let me know, if you get stuck with this architecture.
Thanks, but I am bit confused as to how would joint training and optimization take place.