How to find the total number of parameters in CNN network?

Suppose I am using a deep learning model like VGG-16/ResNet, The other tools like tensorflow, Keras automatically show the number of parameters used by the candidate network.

How do I can find the total number of used paramters if I modfiy the pre-trained network based on my application?

Answer 1

[yanqi liu]

yes，may be use analyzeNetwork to get net model，such as

net = vgg16;
analyzeNetwork(net)

[M M Nabi]

I was actually looking for total number of paramters like model.summary() in tensorflow. Though I know that how many parameters is used in each layer using the designer app like this. L

[Sivylla Paraskevopoulou]

See a similar thread: https://www.mathworks.com/matlabcentral/answers/426886-how-to-calculate-the-number-of-parameters-in-matlab-that-is-used-by-a-deep-learning-network-like-vgg

[Katarina Vuckovic]

analyzeNet shows the total number of learnable parameters per layer in the GUI. Is there a way I can extract that number and input it into the workspace? Right now it seems the only way to caclulate the learnable parameters is to manually sum up all the learnable parameters.

Answer 2

[adel adel]

hello,

try this code:

lgraph = layerGraph(Net);

output = lgraph.Layers(end).Name;

prob = lgraph.Layers(end-1).Name;

lgraph = removeLayers(lgraph,output);

lgraph = removeLayers(lgraph,prob);

dlnet = dlnetwork(lgraph);

numparams = 0;

for i = 1:size(dlnet.Learnables,1)

numparams = numparams + numel(dlnet.Learnables.Value{i});

end

numparams = round(numparams/1000000,3);

end