主要内容

vgg19

VGG-19 convolutional neural network

collapse all in page
  • VGG-19 network architecture

Syntax

NET = VGG19
net = vgg19（'weights'，'imagenet'）
layers = vgg19('Weights','none')

Description

VGG-19 is a convolutional neural network that is 19 layers deep. You can load a pretrained version of the network trained on more than a million images from the ImageNet database[1]。预处理的网络可以将图像分类为1000个对象类别，例如键盘，鼠标，铅笔和许多动物。结果，该网络已经为广泛的图像学习了丰富的功能表示。该网络的图像输入大小为224 by-224。在MATLAB中进行更多预处理的网络®, see预处理的深神经网络。

您可以使用分类to classify new images using the VGG-19 network. Follow the steps ofClassify Image Using GoogLeNet并用VGG-19代替Googlenet。

To retrain the network on a new classification task, follow the steps of训练深度学习网络以对新图像进行分类and load VGG-19 instead of GoogLeNet.

example

网= vgg19returns a VGG-19 network trained on the ImageNet data set.

This function requires Deep Learning Toolbox™ Modelfor VGG-19 Networksupport package. If this support package is not installed, then the function provides a download link.

网= vgg19（'striges'，“ Imagenet”)returns a VGG-19 network trained on the ImageNet data set. This syntax is equivalent toNET = VGG19。

layers= vgg19（'striges'，'none')returns the untrained VGG-19 network architecture. The untrained model does not require the support package.

Examples

collapse all

This example shows how to download and install Deep Learning Toolbox Modelfor VGG-19 Networksupport package.

类型vgg19at the command line.

vgg19

If Deep Learning Toolbox Modelfor VGG-19 Network万博1manbetx未安装支持软件包，然后该功能提供了指向附加探索器中所需的支持软件包的链接。要安装支持包，请单击链接，然后万博1manbetx单击Install。通过键入检查安装成功vgg19at the command line.

vgg19
ans = SeriesNetwork with properties: Layers: [47×1 nnet.cnn.layer.Layer]

Visualize the network using Deep Network Designer.

deepNetworkDesigner(vgg19)

通过单击深度网络设计师中的其他预验证的网络新的。

Deep Network Designer start page showing available pretrained networks

If you need to download a network, pause on the desired network and clickInstallto open the Add-On Explorer.

打开实时脚本

加载预验证的VGG-19卷积神经网络并检查层和类。

利用vgg19加载验证的VGG-19网络。输出网is aSeriesNetwork目的。

NET = VGG19
net =带有属性的系列网络：层：[47×1 nnet.cnn.layer.layer]

View the network architecture using the层s财产。该网络有47层。有19个具有可学习权重的层：16个卷积层和3个完全连接的层。

net.layers
ans = 47x1层阵列，带有图层：1'输入'图像输入224x224x3图像，带有“ zerecenter”标准化2'conv1_1'卷积64 3x3x3卷积，步幅[1 1]和填充[1 1 1 1 1 1 1] 3'relu1_1'relu1_1'relu relu 4 relu relu 4'conv1_2'卷积64 3x3x64大步卷积[1 1]和填充[1 1 1 1] 5'relu1_2'relu 6'pool 6'pool 1'最大ploming 2x2 max poling 2x2 max poling ting stries [2 2]和填充[0 0 0 0 0 0]7'CORV2_1'卷积128 3x3x64大步[1 1]和填充[1 1 1 1] 8'RERU2_1'relu 9'Conv2_2'卷积128 3x3x128速降[1 1]和填充[1 1 1 1 1]10'relu2_2'relu relu 11'pool2'最大池2x2 max plies [2 2]和填充[2 2]和填充[0 0 0 0] 12'conv3_1'卷积256 3x3x128卷积，步幅[1 1]和填充[1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1] 13'relu3_1'relu relu 14'conv3_2'卷积256 3x3x256跨步幅[1 1]和填充[1 1 1 1] 15'relu3_2'relu3_2'relu 16'Conv3_3'Conv3_3'卷积256 3x3x256跨步幅[1 1]和步伐[1]和[1]和填充[1 1 1 1] 17'relu3_3'relu Relu 18'Conv3_4'卷积256 3x3x256步幅[1 1]和填充[1 1 1 1] 19'relu3_4'relu3_4'relu 20'relu 20'pool 20'pool 3 pool 3 pool 3 pool3 pool 3'最大池2x2最大池振步行[2 2]和填充[0 0 0] 21'CORV4_1'卷积512 3X3X256卷积卷积[1 1]和填充[1 1 1 1 1 1 1 1] 22'relu4_1'relu Relu 23'conv4_2'conv4_2'卷积512 3x3x512卷积步伐[1 1]和填充[1 1 1] 24'RERU4_2'RERU RELU RELU 25'CORV4_3'卷积512 3X3X512卷积卷积[1 1]和填充[1 1 1 1 1] 26'relu4_3'relu Relu 27'relu Relu 27'''conv4_4'卷积512 3x3x512大步卷积[1 1]和填充[1 1 1 1] 28'relu4_4'relu relu 29'pool 4 pool4 pool4'最大pl pliming 2x2 max plies plies in tiride [2 2]和填充[0 0 0 0 0] 30 30'conv5_1'卷积512 3x3x512大步卷积[1 1]和填充[1 1 1 1] 31'relu5_1'relu Relu 32'conv5_2'卷积512 3x3x512卷积与步幅[1 1]和填充[1 1 1] 33 'relu5_2' ReLU ReLU 34 'conv5_3' Convolution 512 3x3x512 convolutions with stride [1 1] and padding [1 1 1 1] 35 'relu5_3' ReLU ReLU 36 'conv5_4' Convolution 512 3x3x512 convolutions with stride [1 1] and padding [1 1 1 1] 37 'relu5_4' ReLU ReLU 38 'pool5' Max Pooling 2x2 max pooling with stride [2 2] and padding [0 0 0 0] 39 'fc6' Fully Connected 4096 fully connected layer 40 'relu6' ReLU ReLU 41 'drop6' Dropout 50% dropout 42 'fc7' Fully Connected 4096 fully connected layer 43 'relu7' ReLU ReLU 44 'drop7' Dropout 50% dropout 45 'fc8' Fully Connected 1000 fully connected layer 46 'prob' Softmax softmax 47 'output' Classification Output crossentropyex with 'tench' and 999 other classes

要查看网络学到的类的名称，您可以查看Classes分类输出层的属性（最后一层）。通过指定前10个元素查看前10个类。

net.layers(end).Classes(1:10)
ans =10×1 categorical arraytench goldfish great white shark tiger shark hammerhead electric ray stingray cock hen ostrich

Output Arguments

collapse all

Pretrained VGG-19 convolutional neural network returned as aSeriesNetwork目的。

未经训练的VGG-19卷积神经网络体系结构，以层大批。

参考

[1]ImageNet。http://www.image-net.org

[2] Russakovsky, O., Deng, J., Su, H., et al. “ImageNet Large Scale Visual Recognition Challenge.”International Journal of Computer Vision (IJCV)。Vol 115, Issue 3, 2015, pp. 211–252

[3] Simonyan, Karen, and Andrew Zisserman. "Very deep convolutional networks for large-scale image recognition." arXiv preprint arXiv:1409.1556 (2014).

[4]非常深的卷积网络，用于大规模视觉识别http://www.robots.ox.ac.uk/~vgg/research/very_deep/

Extended Capabilities

Version History

Introduced in R2017a

See Also

||||||||||

Topics