Imagedatastore for image regression, resize image from table

I have used the following link to prepare imagedatasotre for Image to single value regression:

How to prepare imagedatastore for Image to single value regression - MATLAB Answers - MATLAB Central (mathworks.com)

For training:

[net, info] = trainNetwork(table(ims_train.Files,Y_train),lgraph,training_options);

to add validation data:

options = trainingOptions('ValidationData',table(ims_val.Files,Y_val));

However, when I try to train the network I get the following error:

"Error using trainNetwork

The training images are of size 288×382×3 but the input layer expects images of size 256×256×3."

When I create de ImageDatastore I use the 'ReadFcn' option to resize the image.

Is there a way to keep the ReadFcn and not have to rescale all the images in their folders?

Thank you

Answer 1

[Neha]

Hi Iker,

I understand that you want to resize the images to train the model while retaining the original size of the images in the folder. To achieve this, there are alternatives to "ReadFcn" as explained in Preprocess images for Deep Learning. You can refer the code given below to use image augmentation to resize each image on-the-fly during training/validation.

augmenter = imageDataAugmenter();
% variable "imds": imageDataStore
augimdsTrain = augmentedImageDatastore([256,256], imds, 'DataAugmentation', augmenter);  

Hope this helps!