How can I run a trained CNN for classification of a batch of images?

G'day,

I've trained an inception_V3 network using deepNetworkDesigner to classify my images into 2 classes. I've exported this trained network to the workspace and want to now run it on a batch of images.

For these new test images I loaded an entire folder on 100 images onto an imageDatastore object and resized them to 299x299x3 using an augmentedImageDatastore. I needed my newly trained inception_V3 to run on this datastore of 100 images and provide me with labels for each image. I'm guessing the 'predict' or 'classify' commands need to be used but am not able to figure out how to apply these to a datastore.

Also is it possible to know which test image received which prediction label? Is there a way to store the predictions so that the test images can later be separated according to the predictions by the CNN?

Would like to know the syntax/method needed for such batch classification.

Happy to provide more context if required. Thanks!

Answer 1

[V Sairam Reddy]

I understand that you are trying to predict all the test images at once and access their corresponding labels.

Please follow the below mentioned code :

imdsTest = imageDatastore(DatasetPath,IncludeSubfolders=true);

% YTest contains all the predicted labels
YTest = classify(trained_net,imdsTest);

% Accessing Image present at 1st index
index = 1;

% Accessing one Image and its Corresponding label
Image_Data = imread(imdsTest.Files{index});
Predicted_Label = YTest(index);

% Accessing all the Images and corresponding labels
All_Images = readall(imdsTest)
Image_at_Index = All_Images{index};
Corresponding_Label = YTest(index);

Please refer to "Classify Images using Trained Convolutional Neural Network" subsection to know more about "Classify" function and "Specify Images to read" subsection to know more about "Image Datastore" object

[Rishi Ramani]

Thank you very much V for your prompt response. I tried what you suggested and it worked. All 200 images were given classification labels. However, when I tried "Image_at_Index = All_Images[index}; as you suggested it throws up an error "Subscripting into a table using one subscript (as in t(i)) is not supported. Specify a row subscript and a variable subscript, as in t(rows,vars). To select variables, use t(:,i) or for one variable t.(i). To select rows, use t(i,:)." becuase it seems All_Images is a table with two columns (one bieng the size of the image and the other being the label).

But atleast I have all the predicted labels in YTest. Sorry for adding to this question, but now that I have the predicted labels stored in an object, how do I actually separate these images within my PC directory based on the new prediction labels? The goal of my short program is to find out the predicted labels and then separate the classified images into two separate folders based on the prediction labels.

I even tried the following to display the labels and predictions:

numImages = 10;
idx = randperm(numel(auimds.Files),numImages);

figure
tiledlayout("flow")
for i = 1:numImages
    nexttile
    imshow(auimds.Files{idx(i)});
    title("Predicted Label: " + string(YTest(idx(i))))
end

But this just displays 10 random images with their new labels. I want to actually separate these images.

Thank you.

[V Sairam Reddy]

Hey Rishi,

I understand that you are trying to separate the images into two separate folders depending on their prediction labels. Please follow the below mentioned code :

Total_No_Images = numel(auimds.Files);
Total_No_Classes = 2;

% Make 2 folders named 'PredictedAsClass0' and 'PredictedAsClass1'
for i = 0:Total_No_Classes-1
    mkdir(strcat('PredictedAsClass',num2str(i)));
end

% Get the working directory path
working_directory = cd;

% For every image
for i = 1:Total_No_Images
    
    % If prediction is 0, save the image in 'PredictedAsClass0' folder
    if string(YTest(i)) == "0"
        
        % Get the image name
        [filepath,name,ext] = fileparts(auimds.Files{i});
        % Read the image
        image = imread(auimds.Files{i});
        % Save the image with same name in required folder
        imwrite(image,strcat(working_directory, '\', 'PredictedAsClass', string(YTest(i)), '\', name, ext));
    
    end
    
    % Similarly for class 1
    if string(YTest(i)) == "1"
        [filepath,name,ext] = fileparts(auimds.Files{i});
        image = imread(auimds.Files{i});
        imwrite(image,strcat(working_directory, '\', 'PredictedAsClass', string(YTest(i)), '\', name, ext));
    end

end

After the code execution, 2 folders named 'PredictedAsClass0' and 'PredictedAsClass1' are created in the working directory. 'PredictedAsClass0' folder contains all the images whose prediction label is class 0 and vice versa.

[Rishi Ramani]

Hello V,

Thank you very much, you understood me perfectly. Your solution works really well! Instead of that loop to number the folders, I just used mkdir to name the folders of predictions what I wanted.

Thank you for addressing my query and also teaching me a lot about MATLAB!

Wish you the best,

Rishi