NetModel
NetChain
NetDrop
NetGraph
NetInitialize
NetPort
ImageAugmentationLayer
LinearLayer
SoftmaxLayer
DotLayer
MeanSquaredLossLayer
LossFunction

Machine Learning
Image Processing
Computer Vision

Train an Age Estimation Network

Use surgery functionality to train a net from the Wolfram Neural Net Repository

Open in Cloud
Download Notebook

Get an image classification neural net from the repository:

In[1]:=

net=NetModel["ResNet-50 Trained on ImageNet Competition Data"]

Out[1]=

Remove the classification layers and replace them with an age classifier:

In[2]:=

ages=Range[0,100];dims={224,224};ageNet=NetChainImageAugmentationLayerdims,

,NetDrop[net,-2],(*newlayers*)LinearLayer[Length[ages]],SoftmaxLayer[]

Out[2]=

NetChain



uniniti

aliz

Input port:	image
Output port:	vector (size: 101)



Visualize the output of the net with a random initialization. This network estimates an age distribution for any given image:

In[3]:=

ListLinePlotNetInitialize[ageNet][RandomImage[]]//Normal,



Out[3]=

Define a suitable loss function:

In[4]:=

ageLoss=NetGraph[{"ages"ConstantArrayLayer["Array"ages],"meanAge"DotLayer[],"loss"MeanSquaredLossLayer["Input""Real","Target""Real"]},{"ages""meanAge",NetPort["Input"]"meanAge"NetPort[{"loss","Input"}],NetPort["Target"]NetPort[{"loss","Target"}]},"Input"{Length[ages]},"Target""Real"]

Out[4]=

NetGraph



Number of inputs:	2
Loss port:	real



Launch a training using any {image -> age...} training and validation set, like UTKFaces:

In[5]:=

trainingSet=

17,

62,...(*addtrainingdatahere*);validationSet={(*alistofvalidationdata*)}

In[6]:=

trainedNet=NetTrain[ageNet,trainingSet,LossFunctionageLoss,ValidationSetvalidationSet];

Use the trained net to estimate the age of a face in an image:

In[7]:=

face=

;

In[8]:=

Normal[trainedNet[face]].ages//Round

Out[8]=

Plot the distribution of estimated ages:

In[9]:=

ListLinePlottrainedNet[face]//Normal,



Out[9]=

Related Symbols

NetModel
NetChain
NetDrop
NetGraph
NetInitialize
NetPort
ImageAugmentationLayer
LinearLayer
SoftmaxLayer
DotLayer
MeanSquaredLossLayer
LossFunction

Publisher Information

Contributed by: Wolfram Staff

Wolfram Language Example Repository

Train an Age Estimation Network

See Also

Publisher Information