imageImageLayer with variable image size for Fully Convolutional Network (FCN)

Hello all,

I am modifying a pretrained CNN to an FCN and I want to create an input layer that takes variable image dimensions.

layers(1) = imageInputLayer([imageSize 3], 'Name', layers(1).Name,...

'DataAugmentation', layers(1).DataAugmentation, ...

'Normalization', layers(1).Normalization);

Is it possible to initialize imageSize with undefined variables so that it will take automatically the size of each image given in the training process? I tried imageSize = [none none] and other definitions but it is not working.

Appreciate any kind of help!

Thank you