How can I merge (concatenate) feature maps from two different CNN layers in my architecture?

I am looking for something like "merge" layer in Keras to implement U-Net. Thanks!

[Ritesh Chandna]

Unfortunately, we do not have a merge layer feature as of now. However, you can refer to the below link which list all the layer features we have:

https://www.mathworks.com/help/nnet/convolutional-neural-networks.html

Please feel free to provide more information about your workflow.

[Moses Miller]

Hi, Any support for merging LSTM layers in R2018b? Thanks!

Answer 1

[Birju Patel]

Hi,

In R2017b, the depth concatenation layer can be used to merge feature maps:

https://www.mathworks.com/help/nnet/ref/nnet.cnn.layer.depthconcatenationlayer.html

I've also attached an example function that can be used to create U-Net in R2017b (createUnet.m).

HTH, Birju

[Itzik Ben Shabat]

Does this also work for the new 3D layers in 2019a ?

If not, is there an alternative function ?

Answer 2

[o.cefet cefet]

Has anyone ever gotten it on matlab Multiscale Features-Convolutional Neural Network?

Similar to work: https://ieeexplore.ieee.org/document/8625476/