extract features and pass to svm for classification

i want to extract features of my dataset and pass it to svm for classification. when i use pretrain alexnet and do feature extraction then code work well and give me result. but i want to replace pretrained alexnet with mnist dataset (mnist model that i have save already by transfer learning of mnist data from scratch). when i load mnist data and then do feature extraction process then code gives me error ( activations(net,trainingImages,layer); activations(net,testImages,layer); give error in these lines that is in activations). here i attach the code please guide me how to remove the error and what is issue

[Giridharan Kumaravelu]

Could you please post the exact error you get or attach the pretrainedmnist.mat file, so we could know what error you get?

[Alpha Bravo]

To my knowledge, the MNIST data set contains images of the 28 x 28 resolution, while AlexNet was trained on 224 x 224 images. You won't be able to do transfer learning effectively without re-scaling the images.

[amina naseer]

when we rescale images then again it give error

[Thet Hnin Zaw]

hey..i want to also use Alexnet + SVM .So, can i get your coding.

Answer 1

[amina naseer]

Giridharan Kumaravelu sir this is error image

[Giridharan Kumaravelu]

It looks like the loaded pretrained model may not be in the correct format for the written code. Could you please attach the mat files containing the alexnet and pretrainedmnistnet ?

[amina naseer]

Giridharan Kumaravelu sir could you please share your email address so that i will mail files on it. the file is of larger size so that not attached here. please share email

[Giridharan Kumaravelu]

I am not in a position to receive any files. But you could look into this example https://mathworks.com/help/nnet/examples/transfer-learning-using-alexnet.html where it is explained what you are trying to do. As Alpha Bravo mentioned, this could be a dimension mismatch problem. Please check the third dimension of your image too.