-
Notifications
You must be signed in to change notification settings - Fork 2
/
HDA_resnet2vgg_officeHome.m
116 lines (111 loc) · 5.2 KB
/
HDA_resnet2vgg_officeHome.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
% unsupervised domain adaptation
% using LPP
%% Loading Data:
% Features are extracted using resnet101 pretrained on ImageNet without
% fine-tuning
clear all
data_dir = './data/officehome';
domains = {'Art','Clipart','Product','RealWorld'};
num_training_per_class_source = 20;
num_training_per_class_target = 3;
pcaDim = 0;
lppDim = 65;
using_sp = 0;
classifierType = 'nc';
alpha = 10;
lambda = 0.5;
deltaS = 1;
deltaT = 0;
T = 5;
semi = 1; % use semi-supervised learning/unlabelled target data or not?
outfilename = ['./results_semi_HDA/officeHome-resnet2vgg-' classifierType '-sp-' num2str(using_sp) '-alpha-' num2str(alpha) '-lambda' num2str(lambda) '-deltaS-' num2str(deltaS) '-deltaT-' num2str(deltaT) '-PcaDim-' num2str(pcaDim) '-LppDim-' num2str(lppDim) '-T-' num2str(T)]
diary([outfilename '.txt']);
for randseed = 1:10
for source_domain_index = 1:length(domains)
load([data_dir 'OfficeHome-' domains{source_domain_index} '-resnet50-noft.mat']);
feas = resnet50_features;
%load([data_dir 'OfficeHome-' domains{source_domain_index} '-vgg16-noft.mat']);
%feas = vgg16_features; % n * d
domainS_features = L2Norm(feas);
%domainS_features = feas;
domainS_labels = labels+1; % row vector
% dimension reduction for source domain data
if pcaDim >0
options.ReducedDim = pcaDim;
[P_pca,~] = PCA(domainS_features,options);
domainS_features = domainS_features*P_pca;
end
%% training samples selection from source data
rng(randseed);
selector = zeros(1,length(domainS_labels));
for iClass = 1:length(unique(domainS_labels))
numThisClass = sum(domainS_labels==iClass);
if numThisClass <= num_training_per_class_source
selector(domainS_labels==iClass) = 1;
else
randVector= rand(1, numThisClass);
[sorted, sortIndex] = sort(randVector);
selector(domainS_labels==iClass) = randVector<= sorted(num_training_per_class_source);
end
end
trainS_features = domainS_features(logical(selector),:);
trainS_labels = domainS_labels(logical(selector));
for target_domain_index = 1:length(domains)
%if target_domain_index == source_domain_index
% continue;
%end
fprintf('Randseed=%d\n',randseed);
fprintf('Source domain: %s, Target domain: %s\n',domains{source_domain_index},domains{target_domain_index});
%load([data_dir 'OfficeHome-' domains{target_domain_index} '-resnet50-noft.mat']);
%feas = resnet50_features;
load([data_dir 'OfficeHome-' domains{target_domain_index} '-vgg16-noft.mat']);
feas = vgg16_features;
domainT_features = L2Norm(feas);
%domainT_features = feas;
domainT_labels = labels+1;
if pcaDim >0
options.ReducedDim = pcaDim;
[P_pca,~] = PCA(domainT_features,options);
domainT_features = domainT_features*P_pca;
end
%% training and testing samples selection from target data
rng(randseed);
selector = zeros(1,length(domainT_labels));
for iClass = 1:length(unique(domainT_labels))
numThisClass = sum(domainT_labels==iClass);
if numThisClass <= num_training_per_class_target
selector(domainT_labels==iClass) = 1;
else
randVector= rand(1, numThisClass);
[sorted, sortIndex] = sort(randVector);
selector(domainT_labels==iClass) = randVector<= sorted(num_training_per_class_target);
end
end
trainT_features = domainT_features(logical(selector),:);
trainT_labels = domainT_labels(logical(selector));
test_features = domainT_features(logical(1-selector),:);
test_labels = domainT_labels(logical(1-selector));
num_class = length(unique(domainT_labels));
%% HDA with CDLPP
clear options;
options.ReducedDim = lppDim;
options.classifier = classifierType;
options.alpha = alpha;
options.lambda = lambda;
options.NeighborMode = 'KNN';
options.k = 1;
options.WeightMode = 'Cosine';
options.bNormalized = 1;
options.num_iter = T;
options.deltaS = deltaS;
options.deltaT = deltaT;
if using_sp==1
[acc_per_image{source_domain_index}{target_domain_index}{randseed},acc_per_class{source_domain_index}{target_domain_index}{randseed}] = HDA_CDLPP_SP(trainS_features,trainS_labels,trainT_features,trainT_labels,test_features,test_labels,options);
else
[acc_per_image{source_domain_index}{target_domain_index}{randseed},acc_per_class{source_domain_index}{target_domain_index}{randseed}] = HDA_CDLPP(trainS_features,trainS_labels,trainT_features,trainT_labels,test_features,test_labels,options);
end
end
end
end
save([outfilename '.mat'],'acc_per_class*','acc_per_image*');
diary off