Why is activations function not working with custom made CNN model?

3 次查看（过去 30 天）

Jhon Gray 2020-9-30

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/602824-why-is-activations-function-not-working-with-custom-made-cnn-model

I was trying extarct feature from CNN model. I was succesful with series net like alexnet using activation. But it is not wroking with array of CNN layers. Here was my model:

layers = [
    imageInputLayer([32 32 3],"Name","imageinput")
    convolution2dLayer([3 3],512,"Name","conv1_1","WeightsInitializer","narrow-normal")
    tanhLayer("Name","tanh_1")
    maxPooling2dLayer([2 2],"Name","maxpool_2_2","Stride",[2 2])
    convolution2dLayer([5 5],256,"Name","conv1_2","Padding","same","WeightsInitializer","narrow-normal")
    tanhLayer("Name","tanh_2")
    maxPooling2dLayer([2 2],"Name","maxpool_2_1","Stride",[2 2])
    convolution2dLayer([7 7],128,"Name","conv1_3","Padding","same","WeightsInitializer","narrow-normal")
    reluLayer("Name","relu_3")
    maxPooling2dLayer([2 2],"Name","maxpool_3","Stride",[2 2])
    dropoutLayer(0.1,"Name","dropout")
    fullyConnectedLayer(2,"Name","fc","WeightsInitializer","narrow-normal") %Wanna add those features corresponding to each image.
    softmaxLayer("Name","softmax")
    classificationLayer("Name","classoutput")];