-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmymain.m
143 lines (116 loc) · 3.75 KB
/
mymain.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
clear;
addpath(genpath('./'));
nbits_set=[16 32 64 96 128];
%% load dataset
fprintf('loading dataset...\n')
% set = 'CIFAR10';
% set = 'MIRFlickr';
set = 'NUS-WIDE';
% set = 'Places';
if strcmp(set,'MIRFlickr')
load('../Datasets4SelectParam/MIRFLICKR.mat');
I_tr = I_tr(1:18015,:);
L_tr = L_tr(1:18015,:);
elseif strcmp(set,'NUS-WIDE')
load('../Datasets4SelectParam/NUSWIDE10.mat');
I_tr = I_tr(1:40000,:);
L_tr = L_tr(1:40000,:);
elseif strcmp(set,'CIFAR10')
load('../Datasets4SelectParam/cifar10-cut-follow-FOH.mat');
L_tr = L_tr_onehot; L_te = L_te_onehot;
elseif strcmp(set,'Places')
% load('../../Datasets/supervised/Places205_AlexNet_fc7_PCA128');
% L_tr = L_tr_onehot; L_te = L_te_onehot;
opt.dirs.data = '../data_ok/';
DS = datasets.places(opt,0);
trainCNN = DS.Xtrain;
testCNN = DS.Xtest;
trainLabel = DS.Ytrain;
testLabel = DS.Ytest;
test = testCNN ./ sqrt(sum(testCNN .* testCNN, 2));
testLabelvec = full(ind2vec(testLabel')); % c x n
train = trainCNN ./ sqrt(sum(trainCNN .* trainCNN, 2));
trainLabelvec = full(ind2vec(trainLabel')); %c x n
I_tr = train; % 2444772
I_te = test; % 4100
% L_tr = trainLabel;
% L_te = testLabel;
L_tr_onehot = trainLabelvec'; % 2444772 x 205
L_te_onehot = testLabelvec'; % 4100 x 205
L_tr = L_tr_onehot;
L_te = L_te_onehot;
clear trainCNN testCNN trainLabel testLabel train test testLabelvec trainLabelvec L_tr_onehot L_te_onehot opt;
end
anchor=I_tr(randsample(2000,1000),:); %% random select 1000 sample from XTrain (1000*4096)
%% initialization
fprintf('initializing...\n')
if strcmp(set,'MIRFlickr')
% MIR
param.alpha = 1; param.gama = param.alpha;
param.beta = 1;
param.delta = 1;
param.sita = 10;
param.yita = 1;
param.epsilon = 10;
elseif strcmp(set,'NUS-WIDE')
% NUSWIDE
param.alpha = 1; param.gama = param.alpha;
param.beta = 10;
param.delta = 1;
param.sita = 10; param.epsilon = param.sita;
param.yita = 10;
elseif strcmp(set,'CIFAR10')
% CIFAR10
param.alpha = 1; param.gama = param.alpha;
param.beta = 1;
param.delta = 1;
param.sita = 0.1; param.epsilon = param.sita;
param.yita = 100;
end
param.datasets = set;
param.paramiter = 10;
if strcmp(set,'MIRFlickr')
param.nq = 200;
param.n1 = 100;
param.chunk = 2000;
param.nmax = 1000;
elseif strcmp(set,'NUS-WIDE')
param.nq = 400;
param.n1 = 100;
param.chunk = 10000;
param.nmax = 1000;
elseif strcmp(set,'CIFAR10')
param.nq = 200;
param.n1 = 100;
param.chunk = 2000;
param.nmax = 1000;
elseif strcmp(set,'Places')
param.nq = 4000;
param.n1 = 1000;
param.chunk = 100000;
param.nmax = 10000;
end
%% model training
for bit=1:length(nbits_set)
nbits=nbits_set(bit);
Binit = sign(randn(size(I_tr,1), nbits));
Vinit = randn(size(I_tr,1), nbits);
Pinit = randn(1000, nbits);
Sinit = zeros(size(L_tr,2),size(L_tr,2))-1;
param.nbits=nbits;
% randomly generate Teacher codebook
if strcmp(param.datasets,'MIRFlickr')
h = hadamard(512); % 404tags/ 24label
h = h(randperm(size(L_tr,2)),randperm(nbits)); % 404*nbits
elseif strcmp(param.datasets,'NUS-WIDE')
h = hadamard(8192); % 5000
h = h(randperm(size(L_tr,2)),randperm(nbits)); % 5000*nbits
elseif strcmp(param.datasets,'CIFAR10')
h = hadamard(256); % 10
h = h(randperm(size(L_tr,2)),randperm(nbits)); % 5000*nbits
elseif strcmp(param.datasets,'Places')
h = hadamard(256); % 128
h = h(randperm(size(L_tr,2)),randperm(nbits)); % 5000*nbits
end
[ MAP(bit,:),training_time(bit,:)] = train_twostep(I_tr,L_tr,param,I_te,L_te,anchor,Binit,Vinit,Pinit,Sinit,h);
end