matlab 运行 AlexNet

1、alexnet 工具箱下载

下载地址：Neural Network Toolbox(TM) Model for AlexNet Network

    点击下载地址，弹出提示要求先在mathworks注册（已注册的可直接登录），登陆，下载；
    下载完成之后，windows 是无法运行该文件的（alexnet.mlpkginstall）；
    打开 matlab，进入到该文件所在的路径，双击运行；（注：需要较久的时间下载安装 alexnet）
 

2、网络结构：

>> nnet=alexnet;
>> nnet.Layers

ans =

  25x1 Layer array with layers:

     1   'data'     Image Input                   227x227x3 images with 'zerocenter' normalization
     2   'conv1'    Convolution                   96 11x11x3 convolutions with stride [4  4] and padding [0  0  0  0]
     3   'relu1'    ReLU                          ReLU
     4   'norm1'    Cross Channel Normalization   cross channel normalization with 5 channels per element
     5   'pool1'    Max Pooling                   3x3 max pooling with stride [2  2] and padding [0  0  0  0]
     6   'conv2'    Convolution                   256 5x5x48 convolutions with stride [1  1] and padding [2  2  2  2]
     7   'relu2'    ReLU                          ReLU
     8   'norm2'    Cross Channel Normalization   cross channel normalization with 5 channels per element
     9   'pool2'    Max Pooling                   3x3 max pooling with stride [2  2] and padding [0  0  0  0]
    10   'conv3'    Convolution                   384 3x3x256 convolutions with stride [1  1] and padding [1  1  1  1]
    11   'relu3'    ReLU                          ReLU
    12   'conv4'    Convolution                   384 3x3x192 convolutions with stride [1  1] and padding [1  1  1  1]
    13   'relu4'    ReLU                          ReLU
    14   'conv5'    Convolution                   256 3x3x192 convolutions with stride [1  1] and padding [1  1  1  1]
    15   'relu5'    ReLU                          ReLU
    16   'pool5'    Max Pooling                   3x3 max pooling with stride [2  2] and padding [0  0  0  0]
    17   'fc6'      Fully Connected               4096 fully connected layer
    18   'relu6'    ReLU                          ReLU
    19   'drop6'    Dropout                       50% dropout
    20   'fc7'      Fully Connected               4096 fully connected layer
    21   'relu7'    ReLU                          ReLU
    22   'drop7'    Dropout                       50% dropout
    23   'fc8'      Fully Connected               1000 fully connected layer
    24   'prob'     Softmax                       softmax
    25   'output'   Classification Output         crossentropyex with 'tench' and 999 other classes

3、demo（视频与网上研讨会，十一行代码）

deep-learning-in-11-lines-of-matlab-code

clear
camera = webcam;
nnet = alexnet;
while true
    picture = camera.snapshot;
    picture = imresize(picture, [227, 227]);
    label = classify(nnet, picture);
    image(picture);
    title(char(label));
end
 

注意，要运行这个demo，需要安装 MATLAB Support Package for USB Webcams。

请打开附加功能资源管理器以安装 Webcam Support Package