Transfer learning for 2D image using Alexnet

Hello, I am trying to do as following video: https://www.mathworks.com/videos/transfer-learning-with-neural-networks-in-matlab-1492007175092.html

but my data in this case is 2D data, so the network responded some error.

Error using trainNetwork (line 154)
The training images are of size 227x227x1 but the input layer expects images of size 227x227x3.

I have tried to convert the image to 3D image in the file @readFunctionTrain but it does not work (see attached image).

I am not sure about the function @readFunctionTrain but it seems not allow me to do any change.

Have someone ever face this problem?

[Mohammad Sami]

Are your images in grayscale ?

[R G]

yes, they are grayscale

Answer 1

[Mohammad Sami]

Another option would be for you to use the augmented image data store function. It includes both resizing and grey to RGB functionality.

if true
  auimds = augmentedImageDatastore([224,224],imds,'ColorPreprocessing','gray2rgb');
end

https://www.mathworks.com/help/releases/R2018b/deeplearning/ref/augmentedimagedatastore.html?searchHighlight=Resize%20image%20data%20store&s_tid=doc_srchtitle

[R G]

thank you, it works now!

[junyong hong]

sweet af

Answer 2

[Walter Roberson]

You can use a transform() of your datastore https://www.mathworks.com/help/matlab/ref/matlab.io.datastore.transform.html

Or you can use an Augmented Image Datastore, as that has an option to automatically adjust the color.

https://www.mathworks.com/matlabcentral/answers/574165-gan-training-error-with-augmented-image-datastore#answer_474274

[R G]

Im trying your first suggest but get error here, not sure how this code work and what is @(x)

[Mohammad Sami]

The transform function was introduced in R2019a. If you are unable to upgrade, the alternative would be to set a custom ReadFcn function for your imageDatastore