Select random images from each subfolder with ceiling

From a single imagedatastore, I am trying to select 10000 random images from 100 subfolders, each containing 400 to 10000 images, with

each subfolder providing no more than 500 images.

As I read others questions, it looks like I have to

1. Make a new imagedatastore that contains 500 random images from each subfolder (so about 500 images x 100 folders).
2. Select 10000 random images from 1. using randperm().

However, I cannot find a way to do the first task.

Answer 1

[V Sairam Reddy]

I understand that you want a new imageDataStore that contains 500 random images from each of the 100 subfolders.

You can split this task into 2 subtasks :

1. Get a imageDataStore from a single subfolder that contains 500 random images from that subfolder.
2. Combine all such imageDataStores into one big ImageDataStore.

Please follow the below mentioned code :

% Create Image Data Store for a single subfolder.
subFolder1 = fullfile(matlabroot,'toolbox','matlab',{'demos','imagesci'});
exts = {'.jpg','.png','.tif'};
imds_subFolder1 = imageDatastore(subFolder1,'LabelSource','foldernames','FileExtensions',exts);

% Randomly select 500 Images from single subfolder.
total_no_of_Images = length(imds_subFolder1.Files);
RandIndices = randperm(total_no_of_Images);
no_of_reqImages = 500;
% Get first 500 random indices.
indices = RandIndices(1:no_of_reqImages);
% Use 'subset' to extract subset of a DataStore.
imds_subFolder1_with500Images = subset(imds_subFolder1,indices);

% Create Image Data store for another subfolder in similar way.
imds_subFolder2_with500Images = imageDatastore({'street1.jpg','street2.jpg','peppers.png'}); 

% Combine both Image Datastores.
imdsCombined = combine(imds_subFolder2_with500Images,imds_subFolder1_with500Images);



%% Automate for 100 Subfolders

% Place all the 100 different Image Datastores in a cell array.
imds_subFolder = cell(100,1);
for i=1:100
imds_subFolder{i} = imds_ith_subFolder_with500Images;
end

% Combine all the Image Datastores present in the cell array.
CombinedImds = imds_subFolder{1};
for i=2:100
CombinedImds = combine(CombinedImds,imds_subFolder{i});
end

Please refer to Data Store to know more about 'subset' and 'combine' functions, mainly 'Create Subset Datastore with Randomly Selected Files' section in subset documentation.

[Younghoon Kim]

Looks more complicated than I thought but I think I could finally do it.

Thank you!