How do I calculate top5/top1 deep learning errors in MATLAB?

Hello everyone.

I implemented different CNN models (AlexNet, DensNET), and I want to compare them based on top1/top5 error, but I couldn't find any useful tips regarding this point.

Would be appreciated if someone could help me.

Answer 1

[Raynier Suresh]

Top 1 Accuracy:

Output from the model that is the output label with highest probability needs to be same as expected

You can use the below code for Top-1 Accuracy

[YPred,scores] = classify(net,imdsValidation) 
YValidation = imdsValidation.Labels; 
top1Accuracy = mean(YPred == YValidation) 

Top 5 Accuracy:

Any of the top 5 probability label obtained from the network must match with the original label.

You can use the below code for Top-5 Accuracy

[~,scores] = classify(net,imdsValidation); 
YValidation = imdsValidation.Labels;  
[n,m] = size(scores);  
idx = zeros(m,n); 
for i=1:n  
    [~,idx(:,i)] = sort(scores(i,:),'descend');  
end  
idx = idx(1:5,:);  
top5Classes = net.Layers(end).ClassNames(idx);  
top5count = 0;  
for i = 1:n  
    top5count = top5count + sum(YValidation(i,1) == top5Classes(:,i));  
end  
top5Accuracy = top5count/n 

Refer the below links for more information:

https://www.mathworks.com/help/deeplearning/ref/seriesnetwork.classify.html

[Shahad Alqefari]

Thank you so much, I will try it