Training Network for classifying 3D images

[negar noorizadeh]

Dear Sir,

Thanks for sharing this code.

I have 1000 images. each image is 3D image with size [11,11,11].

I have two test images which are 3D images and their size are [11,11,11].

there are 3 class in my problem.

I saw MNIST3d demo and I have some questions and I would be really appreciated if you could help me.

1- I should put 1000 images with size of  [11,11,11] in variable with name of  ''I'' and 1000 label in variable with name of ''Labels'', right?

2- what happened for variables with name of  ''test'' and ''Labels_test''?

3-How can I applied trained network on those two test images with size of  [11,11,11] to define their labels?

Regards

Negar

[Hagay Garty]

Hi Negar - 

1) Yes

2) In your case you are using only 2 images for testing, usually this number should be higher - about 10% of your train set, but if you still wish to use 2 then test should be the same as 'I' but with 2 elements. same goes for labels_test (size 2)

3) See in the doc file, section 5 on how to classify a single image (search for 'classify a single sample') also you can follow the function 'checkNetwork.m' that classify all images in a test set

Regards

[negar noorizadeh]

Dear Hagay,

Thanks for your quick response.

I am a little confused. I still need your help.

I have 1000 images as train set and I have labels of those images.I have also have two unseen images as test images that I don't have their labels.I need to train a classifier and then apply that classifier to my unseen data to predict their labels.

1-So you mean that I should put 900 images and their corresponding labels from train set in variables with names of  ''I'' and ''Labels'' and put  rest of 100 images and their corresponding labels from train set in (10% of train set) in variables with names of  ''test'' and ''Label_test''?right?

2-if yes, it means that, 90% of train set use for training network and 10% for evaluating trained network?

So, how can apply trained network to my unseen data to predict their labels?should I use ''feedForward'' or ''checkNetwork'' ?

Sorry for taking your time and have a great day.

Negar 

[Hagay Garty]

Ok, its more clear now.

1 - Correct

2 - Yes

For classifying a sample you can follow 'checkNetwork' function or read the doc file in section 5 - 'classify a single sample'

Regards,

Hagay

[negar noorizadeh]

Thank you Hagay for all your help