CGAN之条件生成对抗网络（Matlab）
代码来源
代码全文：
clear all; close all; clc;
%% Conditional Generative Adversarial Network
%% Load Data
load('mnistAll.mat')
trainX = preprocess(mnist.train_images);
trainY = mnist.train_labels;
testX = preprocess(mnist.test_images);
testY = mnist.test_labels;
%% Settings
settings.latent_dim = 100;
settings.num_labels = 10;
settings.batch_size = 32; settings.image_size = [28,28,1];
settings.lrD = 0.0002; settings.lrG = 0.0002; settings.beta1 = 0.5;
settings.beta2 = 0.999; settings.maxepochs = 50;
numnodes = 256;
%% Initialization
%% Generator
paramsGen.FCW1 = dlarray(...
initializeGaussian([numnodes,settings.latent_dim],.02));
paramsGen.FCb1 = dlarray(zeros(numnodes,1,'single'));
paramsGen.EMW1 = dlarray(...
initializeUniform([settings.latent_dim,...
settings.num_labels]));
paramsGen.EMb1 = dlarray(zeros(1,settings.num_labels,'single'));
paramsGen.BNo1 = dlarray(zeros(numnodes,1,'single'));
paramsGen.BNs1 = dlarray(ones(numnodes,1,'single'));
paramsGen.FCW2 = dlarray(initializeGaussian(numnodes*[2,1]));
paramsGen.FCb2 = dlarray(zeros(2*numnodes,1,'single'));
paramsGen.BNo2 = dlarray(zeros(2*numnodes,1,'single'));
paramsGen.BNs2 = dlarray(ones(2*numnodes,1,'single'));
paramsGen.FCW3 = dlarray(initializeGaussian(numnodes*[4,2]));
paramsGen.FCb3 = dlarray(zeros(4*numnodes,1,'single'));
paramsGen.BNo3 = dlarray(zeros(4*numnodes,1,'single'));
paramsGen.BNs3 = dlarray(ones(4*numnodes,1,'single'));
paramsGen.FCW4 = dlarray(initializeGaussian(...
[prod(settings.image_size),4*numnodes]));
paramsGen.FCb4 = dlarray(zeros(prod(settings.image_size)...
,1,'single'));
stGen.BN1 = []; stGen.BN2 = []; stGen.BN3 = [];
%% Discriminator
paramsDis.FCW1 = dlarray(initializeGaussian([4*numnodes,...
prod(settings.image_size)],.02));
paramsDis.FCb1 = dlarray(zeros(4*numnodes,1,'single'));
paramsDis.EMW1 = dlarray(...
initializeUniform([prod(settings.image_size),...
settings.num_labels]));
paramsDis.EMb1 = dlarray(zeros(1,settings.num_labels,'single'));
paramsDis.BNo1 = dlarray(zeros(4*numnodes,1,'single'));
paramsDis.BNs1 = dlarray(ones(4*numnodes,1,'single'));
paramsDis.FCW2 = dlarray(initializeGaussian(numnodes*[2,4]));
paramsDis.FCb2 = dlarray(zeros(2*numnodes,1,'single'));
paramsDis.BNo2 = dlarray(zeros(2*numnodes,1,'single'));
paramsDis.BNs2 = dlarray(ones(2*numnodes,1,'single'));
paramsDis.FCW3 = dlarray(initializeGaussian(numnodes*[1,2]));
paramsDis.FCb3 = dlarray(zeros(numnodes,1,'single'));
paramsDis.FCW4 = dlarray(initializeGaussian([1,numnodes]));
paramsDis.FCb4 = dlarray(zeros(1,1,'single'));
stDis.BN1 = []; stDis.BN2 = [];
% average Gradient and average Gradient squared holders
avgG.Dis = []; avgGS.Dis = []; avgG.Gen = []; avgGS.Gen = [];
%% Train
numIterations = floor(size(trainX,2)/settings.batch_size);
out = false; epoch = 0; global_iter = 0;
while ~out
tic;
shuffleid = randperm(size(trainX,2));
trainXshuffle = trainX(:,shuffleid);
trainYshuffle = trainY(shuffleid);
fprintf('Epoch %d\n',epoch)
for i=1:numIterations
global_iter = global_iter+1;
noise = gpdl(randn([settings.latent_dim,...
settings.batch_size]),'CB');
idx = (i-1)*settings.batch_size+1:i*settings.batch_size;
XBatch=gpdl(single(trainXshuffle(:,idx)),'CB');
YBatch=gpdl(single(trainYshuffle(idx)),'B');
[GradGen,GradDis,stGen,stDis] = ...
dlfeval(@modelGradients,XBatch,YBatch,noise,...
paramsGen,paramsDis,stGen,stDis,...
settings);
% Update Discriminator network parameters
[paramsDis,avgG.Dis,avgGS.Dis] = ...
adamupdate(paramsDis, GradDis, ...
avgG.Dis, avgGS.Dis, global_iter, ...
settings.lrD, settings.beta1, settings.beta2);
% Update Generator network parameters
[paramsGen,avgG.Gen,avgGS.Gen] = ...
adamupdate(paramsGen, GradGen, ...
avgG.Gen, avgGS.Gen, global_iter, ...
settings.lrG, settings.beta1, settings.beta2);
if i==1 || rem(i,20)==0
progressplot(paramsGen,stGen,settings);
if i==1 || rem(i,200)==0
h = gcf;
% Capture the plot as an image
frame = getframe(h);
im = frame2im(frame);
[imind,cm] = rgb2ind(im,256);
% Write to the GIF File
if epoch == 0
imwrite(imind,cm,'CGANmnist.gif','gif', 'Loopcount',inf);
else
imwrite(imind,cm,'CGANmnist.gif','gif','WriteMode','append');
end
end
end
end
elapsedTime = toc;
disp("Epoch "+epoch+". Time taken for epoch = "+elapsedTime + "s")
epoch = epoch+1;
if epoch == settings.maxepochs
out = true;
end
end
%% Helper Functions
%% preprocess
function x = preprocess(x)
x = double(x)/255;
x = (x-.5)/.5;
x = reshape(x,28*28,[]);
end
%% extract data
function x = gatext(x)
x = gather(extractdata(x));
end
%% gpu dl array wrapper
function dlx = gpdl(x,labels)
dlx = gpuArray(dlarray(x,labels));
end
%% Weight initialization
function parameter = initializeGaussian(parameterSize,sigma)
if nargin < 2
sigma = 0.05;
end
parameter = randn(parameterSize, 'single') .* sigma;
end
function parameter = initializeUniform(parameterSize,sigma)
if nargin < 2
sigma = 0.05;
end
parameter = 2*sigma*rand(parameterSize, 'single')-sigma;
end
%% Generator
function [dly,st] = Generator(dlx,labels,params,st)
dly = embedding(dlx,labels,params);
% dly = leakyrelu(dly,0.2);
% fully connected
%1
dly = fullyconnect(dly,params.FCW1,params.FCb1);
dly = leakyrelu(dly,0.2);
if isempty(st.BN1)
[dly,st.BN1.mu,st.BN1.sig] = batchnorm(dly,...
params.BNo1,params.BNs1,'MeanDecay',.8);
else
[dly,st.BN1.mu,st.BN1.sig] = batchnorm(dly,params.BNo1,...
params.BNs1,st.BN1.mu,st.BN1.sig,...
'MeanDecay',.8);
end
%2
dly = fullyconnect(dly,params.FCW2,params.FCb2);
dly = leakyrelu(dly,0.2);
if isempty(st.BN2)
[dly,st.BN2.mu,st.BN2.sig] = batchnorm(dly,...
params.BNo2,params.BNs2,'MeanDecay',.8);
else
[dly,st.BN2.mu,st.BN2.sig] = batchnorm(dly,params.BNo2,...
params.BNs2,st.BN2.mu,st.BN2.sig,...
'MeanDecay',.8);
end
%3
dly = fullyconnect(dly,params.FCW3,params.FCb3);
dly = leakyrelu(dly,0.2);
if isempty(st.BN3)
[dly,st.BN3.mu,st.BN3.sig] = batchnorm(dly,...
params.BNo3,params.BNs3,'MeanDecay',.8);
else
[dly,st.BN3.mu,st.BN3.sig] = batchnorm(dly,params.BNo3,...
params.BNs3,st.BN3.mu,st.BN3.sig,...
'MeanDecay',.8);
end
%4
dly = fullyconnect(dly,params.FCW4,params.FCb4);
% tanh
dly = tanh(dly);
end
%% Discriminator
function [dly,st] = Discriminator(dlx,labels,params,st)
dly = embedding(dlx,labels,params);
% dly = leakyrelu(dly,0.2);
% fully connected
%1
dly = fullyconnect(dly,params.FCW1,params.FCb1);
dly = leakyrelu(dly,0.2);
%2
dly = fullyconnect(dly,params.FCW2,params.FCb2);
dly = leakyrelu(dly,0.2);
dly = dropout(dly,.3);
%3
dly = fullyconnect(dly,params.FCW3,params.FCb3);
dly = leakyrelu(dly,0.2);
dly = dropout(dly,.3);
%4
dly = fullyconnect(dly,params.FCW4,params.FCb4);
% sigmoid
dly = sigmoid(dly);
end
%% modelGradients
function [GradGen,GradDis,stGen,stDis]=modelGradients(x,y,z,paramsGen,...
paramsDis,stGen,stDis,settings)
y0 = randi([0,9],[settings.batch_size,1]);
[fake_images,stGen] = Generator(z,y,paramsGen,stGen);
d_output_real = Discriminator(x,y,paramsDis,stDis);
[d_output_fake,stDis] = Discriminator(fake_images,y,paramsDis,stDis);
fake_images0 = Generator(z,y0,paramsGen,stGen);
d_out_fake0 = Discriminator(fake_images0,y0,paramsDis,stDis);
% Loss due to true or not
d_loss = -mean(log(d_output_real+eps)+log(1-d_output_fake+eps));
g_loss = -mean(log(d_out_fake0+eps));
% For each network, calculate the gradients with respect to the loss.
GradGen = dlgradient(g_loss,paramsGen,'RetainData',true);
GradDis = dlgradient(d_loss,paramsDis);
end
%% progressplot
function progressplot(paramsGen,stGen,settings)
r = 2; c = 5;
labels = gpdl(single([0:9]'),'B');
noise = gpdl(randn([settings.latent_dim,r*c]),'CB');
gen_imgs = Generator(noise,labels,paramsGen,stGen);
gen_imgs = reshape(gen_imgs,28,28,[]);
fig = gcf;
if ~isempty(fig.Children)
delete(fig.Children)
end
I = imtile(gatext(gen_imgs));
I = rescale(I);
imagesc(I)
title("Generated Images")
colormap gray
drawnow;
end
%% dropout
function dly = dropout(dlx,p)
if nargin < 2
p = .3;
end
[n,d] = rat(p);
mask = randi([1,d],size(dlx));
mask(mask<=n)=0;
mask(mask>n)=1;
dly = dlx.*mask;
end
%% embedding
function dly = embedding(dlx,labels,params)
% params EM W (latent_dim,num_labels)
% / (img_elements,num_labels)
% b (latent_dim,1) (ignore)
% / (img_elements,1)
maskW = params.EMW1(:,labels+1);
maskb = params.EMb1(:,labels+1);
dly = dlx.*maskW;
% dly = dlx.*maskW+maskb;
end上述代码在Matlab 2020b中运行！！！
代码展示
本代码采用MNIST手写数字数据集（训练集60000个，测试集10000个，本例中采用训练集数据），最大的epoch次数为50,单个epoch中有1875个batch,batch_size为32，结果迭代动图如下：
第1次迭代
第25次迭代
第50次迭代
第75次迭代
第99次迭代
代码源码和论文文献
ck6a
内容简介：
1、上述网址源代码压缩包（本文代码在...\github_repo\CGAN下，CGAN.m文件）
2、条件生成对抗网络的源文献
关注公众号“故障诊断与寿命预测工具箱”，每天进步一点点。