How to find two layers to replace in googlenet?

Hello,

I'm trying the deep learning using googlenet and I don't know how to solve the 'findLayersToReplace'.

I tried this code but it give 3 layers instead of 2 layers that need to find.

layers = net.Layers(end-2:end);

layers = 

    3x1 Layer array with layers:

       1   'loss3-classifier'   Fully Connected         1000 fully connected layer
       2   'prob'               Softmax                 softmax
       3   'output'             Classification Output   crossentropyex with 'tench' and 999 other classes

I don't need to replace Softmax layer.

Please help me creating the function of findLayersToReplace.

Thank you very much

Hana Razak

[houwang]

Hello, I have the same problem. Have you solved your problem?

[mohammed alagele]

hello dear also me the replaceLayer is not work what i do pleas can you help me

Answer 1

[Johannes Bergstrom]

I presume you are looking at this example: https://www.mathworks.com/help/deeplearning/examples/train-deep-learning-network-to-classify-new-images.html

findLayersToReplace is a supporting function/helper function to the example. To access supporting functions of any MATLAB example, open the example by clicking the blue 'Try it in MATLAB' (or similar) button in the top-right of the examples page.

[FEDERICO FURLAN]

Hi Johannes,

Why the softmax layer is not replaced in this example? In other descriptions and examples this layer is always replaced... What is better to do? Thanks

[Chinmay Rane]

Hi the softmax layer is just an activation layer. hence it is not needed to be replaced until you plan to use some other activation. The Fully connected and the classification layer needs your total number of classes, hence we need to replace fc layer and final classification layer(this is set to default as it checks for incoming nodes). Hope it helps

Answer 2

[houwang]

Thank you very much !!! Ihave solved this problems by this function

[marwa za]

hi, please how did you solve your problem ?

[Tahmina Sumi]

@houwang i'm facing the same problem...how did u solve it??

Answer 3

[DEEPA]

TASK

Replace the last fully connected layer of the network with the new layer you just created. The layer that you need to replace is named "loss3-classifier".

Answer 4

[Bhagyashri]

Replace the last fully connected layer of the network with the new layer you just created. The layer that you need to replace is named "loss3-classifier".

Answer 5

[Vedantika]

TASK

Replace the last fully connected layer of the network with the new layer you just created. The layer that you need to replace is named "loss3-classifier".