Train images using DNN using Label as CSV file

I want to train fundus images (available at kaggle websie for diabetic retinopathy) using pretrained neural network. Dataset consist of images and labels are available in csv file. Kindly guide how can i load images and provide lables as csv file in matlab. Waiting for seniors favorable reply

Answer 1

[Mohammad Sami]

From your question it seems like you want to do transfer learning. Matlab documentation provide a detailed guide on how to do transfer learning. You can check it out here.

https://www.mathworks.com/help/deeplearning/gs/get-started-with-transfer-learning.html

If you are new to this area I will also suggest you to do free online training provided here:

https://www.mathworks.com/learn/tutorials/deep-learning-onramp.html

[Farrukh nazir]

Thanks for reply however i was asking that how can i provide labels as .csv file in dnn for checking validation data E.g as in below mentioned matlab command, label source is fildernames but i want to give .csv file as lable source [Imdstrain, imdsvalidation]=imageDatastore(filepath, 'includesubfolders', true,'labelsource','foldername')

[Mohammad Sami]

You will have to manually set the Labels property in that case.

Make sure your Labels are in the same order as Files property.

The best way would be your csv contains the Files as one column.

% labelsdata = readtable('labels.csv');
[match,index] = ismember(labelsdata.Files,imds.Files);
if ~all(match)
    warning('Not all the files have labels');
end
imds.Labels(match) = labelsdata.Labels(index(match));