Hi, Is it possible to train a Convolutional Neural Network for NxNxM inputs, where M is the size of the channel and is greater than 3?

 The Image Input layer of a convolutional neural network only takes in NxNx1 (gray scale) or NxNx3 (RGB). However my data is NxNx25, is there a way to train this sort of data with Matlab?

Answer 1

[Sindhu Yerragunta]

Hi Vince,

As of now it is not possible to train a convolutional Neural Network to be able to work with more than 3 channels data.

This enhancement will be supported in one of the future releases of Neural Network Toolbox.

-Sindhu

[vince65201]

Thanks! I thought I missed out something.

[Vamshi Reddy]

Is it possible in 2019a or 2019b???. Iam using 2018b.

I was unable to pass more than 3 channels from one layer to another layer which is a custom layer defined by me.

When I tried to pass 128 channels ,it is giving an error :

Error using builtin

Array inputs must share the same number of channels.

can u help me to fix this??? @Sindhu Yerragunta

[Pedapudi Bharath Raja Bhoopal]

@Sindhu Yerragunta

Can you please comfirm if multi channel is supported in latest matlab desktop version? (name the version)

I have observed that in matlab online, multichannel is acceptable for ImageInputLayer function.

inLayer=imageInputLayer([28 28 128])

inLayer = 

  ImageInputLayer with properties:

                      Name: ''
                 InputSize: [28 28 128]

   Hyperparameters
          DataAugmentation: 'none'
             Normalization: 'zerocenter'
    NormalizationDimension: 'auto'
                      Mean: []

Answer 2

[Zhiyi TANG]

Check the doc of imageInputLayer (2018a) below:

https://www.mathworks.com/help/nnet/ref/nnet.cnn.layer.imageinputlayer.html?s_tid=srchtitle