-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathunsupervised_DA_SP_imageclef.m
71 lines (70 loc) · 3.6 KB
/
unsupervised_DA_SP_imageclef.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
% =====================================================================
% Code for conference paper:
% Qian Wang, Toby Breckon, Unsupervised Domain Adaptation via Structured Prediction Based Selective Pseudo-Labeling, AAAI2020
% By Qian Wang, [email protected]
% =====================================================================
%% Loading Data:
% Features are extracted using resnet50 pretrained on ImageNet without
% fine-tuning
clear all %#ok<CLALL>
addpath('./utils/');
data_dir = '/mnt/HD2T/DomainAdaptation/image-clef/';
%data_dir = 'E:\DomainAdaptation\Office31data\';
domains = {'c','i','p'};
count = 0;
for source_domain_index = 1:length(domains)
load([data_dir 'imageclef-' domains{source_domain_index} '-resnet50-noft']);
domainS_features_ori = L2Norm(resnet50_features);
%domainS_features_ori = resnet50_features;
domainS_labels = labels+1;
for target_domain_index = 1:length(domains)
if target_domain_index == source_domain_index
continue;
end
fprintf('Source domain: %s, Target domain: %s\n',domains{source_domain_index},domains{target_domain_index});
load([data_dir 'imageclef-' domains{target_domain_index} '-resnet50-noft']);
domainT_features = L2Norm(resnet50_features);
%domainT_features=resnet50_features;
opts.ReducedDim = 128;
X = double([domainS_features_ori;domainT_features]);
P_pca = PCA(X,opts);
domainS_features = domainS_features_ori*P_pca;
domainT_features = domainT_features*P_pca;
domainS_features = L2Norm(domainS_features);
domainT_features = L2Norm(domainT_features);
domainT_labels = labels+1;
num_class = length(unique(domainT_labels));
%% Baseline method: using 1-NN, only labelled source data for training
% fprintf('Baseline method using 1NN:\n');
% classifierType='1nn';
% [acc_per_img,acc_per_class]= func_recognition(domainS_features,domainT_features,domainS_labels,domainT_labels,classifierType);
% count = count+1;
% acc1nn(count) = acc_per_img;
% %% Baseline method: using NC, only labelled source data for training
% % fprintf('Baseline method using NC:\n');
% % classifierType='nc';
% % acc= func_recognition(domainS_features,domainT_features,domainS_labels,domainT_labels,classifierType);
% %% Baseline method: using SVM, only labelled source data for training
% fprintf('Baseline method using SVM:\n');
% classifierType='svm';
% [acc_per_img, acc_per_class] = func_recognition(domainS_features,domainT_features,domainS_labels,domainT_labels,classifierType);
% accsvm(count)=acc_per_img;
%% Proposed method:
fprintf('Proposed method using 1NN:\n');
using_sp = 1;
d = 128;
T = 11;
if using_sp
[ acc,acc_per_class ]= DA_LPP_SP(domainS_features,domainS_labels,domainT_features,domainT_labels,d,T);
else
[ acc,acc_per_class ]= DA_LPP(domainS_features,domainS_labels,domainT_features,domainT_labels,d,T);
%acc_per_class = DA_LDA(domainS_features,domainS_labels,domainT_features,domainT_labels);
end
count = count + 1;
all_acc_per_class(count,:) = mean(acc_per_class,2);
all_acc_per_image(count,:) = acc;
end
end
mean_acc_per_class = mean(all_acc_per_class,1)
mean_acc_per_image = mean(all_acc_per_image,1)
save(['imageclef-SP-' num2str(using_sp) '-PcaDim-' num2str(opts.ReducedDim) '-LppDim-' num2str(d) '-T-' num2str(T) '.mat'],'all_acc_per_class','all_acc_per_image','mean_acc_per_class','mean_acc_per_image');