How to create a multiple-branch CNN model?

Dear Coleagues,

I am trying to create a double-branch CNN, where each branch will receive as input a different image (i.e., each branch has its own 'InputLayer'). I want these two branches (with their layers) to be merged at the end, by means of a common fully-connected-layer. I was trying to implement it using the DAG Network. However, when using addLayers function, it gives me an error saying Layer graphs can contain only one input layer.

Is there any other way to create multiple-branch CNNS in Matlab? Any possibility to use DAG Networks receiving more than one inputLayer?

Thank you all. Ezequiel.

Answer 1

[Kenta]

https://jp.mathworks.com/matlabcentral/fileexchange/74760-image-classification-using-cnn-with-multi-input-cnn

Hi, I think what you would like to do is shown above.

Other CNN architechture examples that accept multi-input layers were below. Hope it helpes you.

https://jp.mathworks.com/matlabcentral/fileexchange/75441-conditional-gan-and-cnn-classification-with-fashion-mnist

https://jp.mathworks.com/matlabcentral/fileexchange/74974-conditional-vae-variational-auto-encoder-vae