What is the correct syntax for using an augmentedImageDatastore as validation data in trainingOptions for the trainNetwork() function?

I have the following image datastore:
myDataset = imageDatastore('C:\MyData',...
'IncludeSubfolders',true,...
'FileExtensions','.png',...
'LabelSource','foldernames');
[imdsTrain, imdsVal, imdsTest] = splitEachLabel(myDataset, 0.6, 0.2, 'randomized');
Which I then prepare to input into some pre-trained networks:
augimdsTrain = augmentedImageDatastore([227 227 3],imdsTrain,...
'ColorPreprocessing', 'gray2rgb',...
'DataAugmentation', augmenter);
augimdsVal = augmentedImageDatastore([227 227 3],imdsVal, ...
'ColorPreprocessing', 'gray2rgb');
augimdsTest = augmentedImageDatastore([227 227 3],imdsTest,...
'ColorPreprocessing', 'gray2rgb');
What is the correct syntax for using augimdsVal as 'ValidationData'in trainingOptions? Do I just use the augmented datastore as is (as I've seen in some of the documentation):
options = trainingOptions('sgdm','InitialLearnRate', 0.0001, 'ValidationData', augimdsVal);
Or can I add labels like so:
options = trainingOptions('sgdm','InitialLearnRate', 0.0001, 'ValidationData', {augimdsVal imdsVal.Labels});
Or is there a better way?

 Accepted Answer

Hi,
The name-value pair 'ValidationData' can support only the following types as a value:
  1. Image Data Store.
  2. Data Store.
  3. Table.
  4. Cell array of X and Y, where X represents data while Y represents corresponding labels.
So, I guess it is invalid to give labels again as a value.
Refer to the following link:
Hope this clarifies your concern!

3 Comments

Thanks for the update. I was a little confused as some of the documentation examples such as 'Transfer Learning using AlexNet' use an augmentedImageDatastore as the 'ValidationData' and I was wondering how could validation occur without the labels.
As the imagedataStore has labels, there is no need to explicitly specify labels to augmented image datastore, when we use imds as the argument during creation.
Great stuff. That's exactly what I wanted to know. Thanks.

Sign in to comment.

More Answers (0)

Categories

Find more on Deep Learning Toolbox in Help Center and File Exchange

Products

Release

R2020a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!