I am trying to find deep learned features at fc7 layer of alexnet

I am trying to find deep learned features at fc7 layer of alexnet using the below code:

close all;

clear all;

clc;

run './matconvnet/matlab/vl_setupnn'

% Load pre-trained AlexNet

net = alexnet;

% Load an example image

im = imread('01.jpg');

% Resize the image to the input size of AlexNet

im = imresize(im, [227 227]);

% Preprocess the image for AlexNet

im = im2single(im);

%im = im - net.meta.normalization.averageImage;

for j=1:3

im(:,:,j) = im(:,:,j) - net.meta.normalization.averageImage(j);

end

% Extract features at fc7 layer

features = activations(net, im, 'fc7');

% Display the extracted features

disp(features);

I am getting this error:

Unrecognized method, property, or field 'meta' for class 'SeriesNetwork'.

Error in alexnet_fc7 (line 19)

im(:,:,j) = im(:,:,j) - net.meta.normalization.averageImage(j);

Answer 1

[Himanshu]

Hello Aditya,

As per my understanding, you are facing an error while extracting the features from the "fc7" Fully Connected Layer of the AlexNet Convolutional Neural Network.

The error occurs because you are trying to access the "meta" field of the "net" object, which doesn't exist in the "SeriesNetwork" class.

As per the provided code, I understand that you want to normalize the input data by subtracting the mean of the dataset images. You can perform the operation by replacing the "for" loop with the following code:

average_image = mean(net.Layers(1).Mean(1,:), 2)
im = bsxfun(@minus, im, average_image)

Here, it first calculates the average image value using the "Mean" property of the first layer and then subtracts it from the input image.

You can refer to the below documentation to understand more about the "AlexNet", "SeriesNetwork" and "bsxfun" function.

AlexNet convolutional neural network - MATLAB alexnet - MathWorks India

Series network for deep learning - MATLAB - MathWorks India

Apply element-wise operation to two arrays with implicit expansion enabled - MATLAB bsxfun - MathWorks India

[Aditya]

Thanks a lot Himanshu, This method is working and I am getting the output for the same.