My trained network only shows 4 layers, Why can't I see the whole architecture?

Question

Laraib 2023-7-6

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1992608-my-trained-network-only-shows-4-layers-why-can-t-i-see-the-whole-architecture

回答： Jayant 2023-7-6

I am using following code to train inception V3 on my data

fpath='E:\Research Data\thesis';

data=fullfile(fpath,'E229A');

tdata=imageDatastore(data,'IncludeSubfolders',true,'LabelSource','foldernames');

net= inceptionv3;

layers=[imageInputLayer([229 229 3])

net(2:end-3)

fullyConnectedLayer(2)

softmaxLayer

classificationLayer

];

opt=trainingOptions("sgdm","MaxEpochs",30,'InitialLearnRate',0.001);

Incepv3=trainNetwork(tdata,layers,opt);

save('Incepv3.mat', 'Incepv3');

But when I load the trained network it only shows 4 layers as follows

load('Incepv3.mat', 'Incepv3');

>> fullLayers = Incepv3.Layers;

>> disp(fullLayers);

4×1 Layer array with layers:

1 'imageinput' Image Input 229×229×3 images with 'zerocenter' normalization

2 'fc' Fully Connected 5 fully connected layer

3 'softmax' Softmax softmax

4 'classoutput' Classification Output crossentropyex with 'BrownP' and 1 other classe

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Jayant 2023-7-6

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1992608-my-trained-network-only-shows-4-layers-why-can-t-i-see-the-whole-architecture#answer_1268533

在 MATLAB Online 中打开

It seems that you are fine-tuning the Inception V3 network by replacing the last few layers with your own layers for classification. This modification might have caused the discrepancy in the displayed layers.

To verify the complete architecture, you can try accessing the layers of the original Inception V3 network

net = inceptionv3;
disp(net.Layers);

If you have made modifications to the network structure, it's possible that the displayed layers might differ. In that case, you can manually inspect the layers you have defined in your code to ensure that the desired modifications have been applied correctly.

To see the last three modified layers along with the rest of the network architecture, you can combine the layers from the loaded network (Incepv3) with the layers of the original Inception V3 network (net).

load('Incepv3.mat', 'Incepv3');
net = inceptionv3;
% Get the modified layers from Incepv3
modifiedLayers = Incepv3.Layers(end-2:end);
% Combine the modified layers with the rest of the network layers
fullLayers = [net.Layers; modifiedLayers];
% Display the complete network architecture
disp(fullLayers);