Is it possible to use pretrained network like alexnet with a new dataset that has different category?

Question

Suheer Ali 2018-12-13

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/435502-is-it-possible-to-use-pretrained-network-like-alexnet-with-a-new-dataset-that-has-different-category

回答： Mahmoud Afifi 2020-1-6

I used alexnet with a dataset which it is categores {1-50,51-100, 101- ...}. is this possible?

if yes, why I get this error?

Error using trainNetwork (line 150)
Training and validation data must have the same labels.
Error in AlexNet_ex_density (line 114)
netTransfer = trainNetwork(augimdsTrain,layers,options);
Caused by:
    Error using trainNetwork>iAssertClassNamesAreTheSame (line 376)
    Training and validation data must have the same labels.

I used

[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomized');
% Create augmentedImageDatastore from training and test sets to resize
% images in imds to the size required by the network.
imageSize = net.Layers(1).InputSize;
augmentedTrainingSet = augmentedImageDatastore(imageSize, imdsTrain, 'ColorPreprocessing',...
    'gray2rgb','DataAugmentation',imageAugmenter);
augmentedTestSet = augmentedImageDatastore(imageSize, imdsTest, 'ColorPreprocessing', 'gray2rgb',...
    'DataAugmentation',imageAugmenter);
augimdsValidation = augmentedImageDatastore(imageSize, imdsValidation, 'ColorPreprocessing',...
    'gray2rgb','DataAugmentation',imageAugmenter);

and in options of the network I used

'ValidationData',augimdsValidation, ...
    'ValidationFrequency',3, ...

then I uesd

netTransfer = trainNetwork(augimdsTrain,layers,options);

I got the error in above sentince.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Mahmoud Afifi 2020-1-6

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/435502-is-it-possible-to-use-pretrained-network-like-alexnet-with-a-new-dataset-that-has-different-category#answer_408864

You need to change the last fully connected layer of Alexnet with a new one with the same number of expected output (either for regression or number of classes for classification).

https://www.mathworks.com/help/deeplearning/examples/transfer-learning-using-alexnet.html;jsessionid=82bda7d7b633139e3e0a8f6e685d