Unable to perform assignment because the size of the left side is 1-by-4680 and the size of the right side is 1-by-9360, how can i solve this error?

%%Extract HOG Features for training set 
trainingFeatures = zeros(size(training,2)*training(1).Count,4680);
featureCount = 1;
for i=1:size(training,2)
    for j = 1:training(i).Count
        trainingFeatures(featureCount,:) = extractHOGFeatures(read(training(i),j));
        trainingLabel{featureCount} = training(i).Description;    
        featureCount = featureCount + 1;
    end
    personIndex{i} = training(i).Description;
end

[Bob Thompson]

Which line gives the error?

[Walter Roberson]

What size is your image?

I notice that the output size is exactly twice what you were expecting.

[karim ghzaiel]

@Walter Roberson @salma salah

can you explain me why we choose 4680

4680 refering to what ? thank you

[Walter Roberson]

https://www.mathworks.com/help/vision/ref/extracthogfeatures.html#btxscw9-features

gives the formula

[Shefali Patil]

same code i have in my project and i m getting error in this line :

trainingFeatures(featureCount,:) = extractHOGFeatures(read(training(i),j));

i am not getting what changes to do here, what to type instead of 4680 ? how can we say what is the size of every image ?

Answer 1

[Walter Roberson]

According to the documentation, the returned size is https://www.mathworks.com/help/vision/ref/extracthogfeatures.html#btxscw9-features

HOG feature length, N, is based on the image size and the function parameter values.

N = prod([BlocksPerImage, BlockSize, NumBins])

BlocksPerImage = floor((size(I)./CellSize – BlockSize)./(BlockSize – BlockOverlap) + 1)

The cell size defaults to 8 x 8 according to https://www.mathworks.com/help/vision/ref/extracthogfeatures.html#btxscw9-CellSize

The BlockSize defaults to 2 x 2 according to https://www.mathworks.com/help/vision/ref/extracthogfeatures.html#btxscw9-BlockSize

Numbins defaults to 9 according to https://www.mathworks.com/help/vision/ref/extracthogfeatures.html#btxscw9-NumBins

BlockOverlap defaults to ceil(BlockSize/2) according to https://www.mathworks.com/help/vision/ref/extracthogfeatures.html#btxscw9-BlockOverlap ; with the default BlockSize of 2 x 2 that would be a default of 1 x 1

So we can calculate

%example image
I = imread('flamingos.jpg');

%the calculation
BlocksPerImage = floor(([size(I,1), size(I,2)] ./ 8 - 2) ./ (8 - 1) + 1)

BlocksPerImage = 1×2

18 23

and then

N = prod([BlocksPerImage, [8 8], 9])

N = 238464

Now let us try it:

output = extractHOGFeatures(I);
size(output)

ans = 1×2

1 695520

The difference suggests a factor of 3 due to 3 color planes. I will investigate.

[Walter Roberson]

Unfortunately the routine that does the final computation is a built-in so I cannot determine exactly where the difference is coming from.

I will file a request to have the documentation expanded for the truecolor case.