/
LSGAN.m
247 lines (230 loc) · 7.51 KB
/
LSGAN.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
clear all; close all; clc;
%% Least Squares 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.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;
%% Initialization
%% Generator
paramsGen.FCW1 = dlarray(...
initializeGaussian([256,settings.latent_dim],.02));
paramsGen.FCb1 = dlarray(zeros(256,1,'single'));
paramsGen.BNo1 = dlarray(zeros(256,1,'single'));
paramsGen.BNs1 = dlarray(ones(256,1,'single'));
paramsGen.FCW2 = dlarray(initializeGaussian([512,256]));
paramsGen.FCb2 = dlarray(zeros(512,1,'single'));
paramsGen.BNo2 = dlarray(zeros(512,1,'single'));
paramsGen.BNs2 = dlarray(ones(512,1,'single'));
paramsGen.FCW3 = dlarray(initializeGaussian([1024,512]));
paramsGen.FCb3 = dlarray(zeros(1024,1,'single'));
paramsGen.BNo3 = dlarray(zeros(1024,1,'single'));
paramsGen.BNs3 = dlarray(ones(1024,1,'single'));
paramsGen.FCW4 = dlarray(initializeGaussian(...
[prod(settings.image_size),1024]));
paramsGen.FCb4 = dlarray(zeros(prod(settings.image_size)...
,1,'single'));
stGen.BN1 = []; stGen.BN2 = []; stGen.BN3 = [];
%% Discriminator
paramsDis.FCW1 = dlarray(initializeGaussian([1024,...
prod(settings.image_size)],.02));
paramsDis.FCb1 = dlarray(zeros(1024,1,'single'));
paramsDis.BNo1 = dlarray(zeros(1024,1,'single'));
paramsDis.BNs1 = dlarray(ones(1024,1,'single'));
paramsDis.FCW2 = dlarray(initializeGaussian([512,1024]));
paramsDis.FCb2 = dlarray(zeros(512,1,'single'));
paramsDis.BNo2 = dlarray(zeros(512,1,'single'));
paramsDis.BNs2 = dlarray(ones(512,1,'single'));
paramsDis.FCW3 = dlarray(initializeGaussian([256,512]));
paramsDis.FCb3 = dlarray(zeros(256,1,'single'));
paramsDis.FCW4 = dlarray(initializeGaussian([1,256]));
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;
trainXshuffle = trainX(:,randperm(size(trainX,2)));
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');
[GradGen,GradDis,stGen,stDis] = ...
dlfeval(@modelGradients,XBatch,noise,...
paramsGen,paramsDis,stGen,stDis);
% 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 || (epoch>=0 && i==1)
% 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,'GANmnist.gif','gif', 'Loopcount',inf);
% else
% imwrite(imind,cm,'GANmnist.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
%% Generator
function [dly,st] = Generator(dlx,params,st)
% fully connected
%1
dly = fullyconnect(dlx,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);
else
[dly,st.BN1.mu,st.BN1.sig] = batchnorm(dly,params.BNo1,...
params.BNs1,st.BN1.mu,st.BN1.sig);
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);
else
[dly,st.BN2.mu,st.BN2.sig] = batchnorm(dly,params.BNo2,...
params.BNs2,st.BN2.mu,st.BN2.sig);
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);
else
[dly,st.BN3.mu,st.BN3.sig] = batchnorm(dly,params.BNo3,...
params.BNs3,st.BN3.mu,st.BN3.sig);
end
%4
dly = fullyconnect(dly,params.FCW4,params.FCb4);
% tanh
dly = tanh(dly);
end
%% Discriminator
function [dly,st] = Discriminator(dlx,params,st)
% fully connected
%1
dly = fullyconnect(dlx,params.FCW1,params.FCb1);
dly = leakyrelu(dly,0.2);
dly = dropout(dly);
if isempty(st.BN1)
[dly,st.BN1.mu,st.BN1.sig] = batchnorm(dly,params.BNo1,params.BNs1);
else
[dly,st.BN1.mu,st.BN1.sig] = batchnorm(dly,params.BNo1,...
params.BNs1,st.BN1.mu,st.BN1.sig);
end
%2
dly = fullyconnect(dly,params.FCW2,params.FCb2);
dly = leakyrelu(dly,0.2);
dly = dropout(dly);
if isempty(st.BN2)
[dly,st.BN2.mu,st.BN2.sig] = batchnorm(dly,params.BNo2,params.BNs2);
else
[dly,st.BN2.mu,st.BN2.sig] = batchnorm(dly,params.BNo2,...
params.BNs2,st.BN2.mu,st.BN2.sig);
end
%3
dly = fullyconnect(dly,params.FCW3,params.FCb3);
dly = leakyrelu(dly,0.2);
dly = dropout(dly);
%4
dly = fullyconnect(dly,params.FCW4,params.FCb4);
% sigmoid
% dly = sigmoid(dly);
end
%% modelGradients
function [GradGen,GradDis,stGen,stDis]=modelGradients(x,z,paramsGen,...
paramsDis,stGen,stDis)
[fake_images,stGen] = Generator(z,paramsGen,stGen);
d_output_real = Discriminator(x,paramsDis,stDis);
[d_output_fake,stDis] = Discriminator(fake_images,paramsDis,stDis);
% Loss due to true or not
d_loss = .5*mean((d_output_real-1).^2+(d_output_fake).^2);
g_loss = .5*mean((d_output_fake-1).^2);
% 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 = 5; c = 5;
noise = gpdl(randn([settings.latent_dim,r*c]),'CB');
gen_imgs = Generator(noise,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 = p*10;
mask = randi([1,10],size(dlx));
mask(mask<=n)=0;
mask(mask>n)=1;
dly = dlx.*mask;
end