-
Notifications
You must be signed in to change notification settings - Fork 0
/
ntu_cross_view_mining_new_vlad.m
503 lines (431 loc) · 17.5 KB
/
ntu_cross_view_mining_new_vlad.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
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
clear all;
addpath(genpath('liblinear-2.1'));
addpath('general_function\');
addpath('libraryPubliced\tool\liblinear\');
addpath('libraryPubliced\tool\liblinear\matlab\');
%addpath('libraryPubliced\tool\libsvm\matlab\');
run('I:\wangyancheng\code\vlfeat-0.9.21\toolbox\vl_setup');
clusterfun = 'no'; % or 'dplim', or 'no', or 'kmlim';
numClass = 60;
numCluster_list = [30 30 30 30 30 30 30 30 30 ];
numClSelect = 2;
numC = 8;
numBlock = 1;
gmmSamRatio = 0.6;
% seg_c = [1 2 3 4 5];
seg_c = [1 2 3 4 5];
isPCA = 0;
isFVNorm = 0;
isFV = 0;
isMerge =0;
path = 'G:\wangyancheng\ntu_rgbd\dynamic\feature_ntu_view_share_MBB\';% data path
index_s = ['01';'02';'03';'04';'05';'06';'07';'08';'09';'10';'11';'12';'13';'14';'15';'16';'17'];
index_c = ['01';'02';'03'];
index_p = ['01';'02';'04';'05';'08';'09';'13';'14';'15';'16';'17';'18';'19';'25';'27';'28';'31';'34';'35';'38';...
'03';'06';'07';'10';'11';'12';'20';'21';'22';'23';'24';'26';'29';'30';'32';'33';'36';'37';'39';'40'];%
index_r = ['01';'02'];
index_a = ['01';'02';'03';'04';'05';'06';'07';'08';'09';'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'];
%data_train = zeros(4096,5,11,length(dir([path '*_v01*']))+length(dir([path '*_v02*'])));
dimension = 4096;
if isPCA dPCA = 1024; else dPCA=dimension; end
sizeBlock = dPCA/numBlock;
new_path1 = 'D:\wangyancheng\code\ntudata';
if 1
%--------- extract the data
flag = [];
data_train = [];
data_test = [];
idx_trn =1;
idx_tst = 1;
lab_trn = [];
lab_tst = [];
fileNames_Tr = [];
if ~exist(new_path1)
mkdir(new_path1); % 若不存在，在当前目录中产生一个子目录‘Figure’
end
fd= fopen('save_name.txt', 'w');
%--------- extract the data
for i = 1:length(index_s)
i
for j = 1:length(index_c)
for k = 1:length(index_p)
for m =1:length(index_r)
for d = 1:length(index_a)%-50
filename = ['S0',index_s(i,:), 'C0' ,index_c(j,:),'P0',index_p(k,:),'R0',index_r(m,:),'A0',index_a(d,:),'.mat'];
if exist([path filename],'file')
if j>1
fprintf(fd,'%s\r\n',filename);
fileNames_Tr = [fileNames_Tr;filename];
dt_train1 = load([path filename]);
data = reshape(cell2mat(dt_train1.data),[4096,5,11,2]);
data_train(:,:,:,idx_trn) = squeeze(data(:,:,:,1));
idx_trn = idx_trn+1;
lab_trn = [lab_trn;d];
% in order to split the trainset
if(k>35)
flag = [flag;1];
else
flag = [flag;0];
end
else
dt_test = load([path filename]);
data = reshape(cell2mat(dt_test.data),[4096,5,11,2]);
data_test(:,:,:,idx_tst) = squeeze(data(:,:,:,1));
lab_tst = [lab_tst;d];
idx_tst = idx_tst+1;
end
end
end
end
end
end
end
fclose(fd);
end
%------- PCA --------%
% if isPCA
% fprintf('----PCA----\n');
% dt = reshape(data_train,[4096 5*11*size(data_train,4)]);
% [pc,score,latent,tsquare] = pca(dt');
% after_pca = score(:,1:dPCA);
%
% dt = reshape(data_test,[4096 5*11*size(data_test,4)]);
% test_pca = dt' * pc(:,1:dPCA);
% sample.train = reshape(after_pca',[dPCA 5 11 numSam_train]);
% sample.test = reshape(test_pca',[dPCA 5 11 numSam_test]);
%
% end
numSam_train = size(data_train,4);
numSam_test = size(data_test,4);
size(data_train)
size(data_test)
% sample.all = cat(4,data_train,data_test); % help cat [4096 5 11 N]
%
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% for i=1:size(data_train,2)
%
% filename1 = ['ntu_datamat_',num2str(i),'.mat'];
% if ~exist([new_path1 '\' filename1])
%
% sub_sample = cat(3,squeeze(data_train(:,i,:,:)),squeeze(data_train(:,i,:,:)));
% save([new_path1 '\' filename1],'sub_sample','lab_trn','numSam_train','numSam_test','flag','lab_tst','-v7.3');
% end
% end
% clear data_train;
%
% clear data_test;
% end
% merge for segmation
se = 5;
%sample = Normlize(sample);
samp = [];
% if isMerge
% for p = 2:4
% for d = (p+1):5
% p_max = sample.train(:,d,:,:)-sample.train(:,p,:,:);
% sample.train = [sample.train p_max];
% p_max = sample.test(:,d,:,:)-sample.test(:,p,:,:);
% sample.test = [sample.test p_max];
% end
% end
% se=9;
% end
% sample.train = samp(:,:,:,1:numSam_train);
% sample.test = samp(:,:,:,numSam_train+1:end);
% sample.all = [];
% sample.all = samp;
%
% sample.train = sample.all(:,:,:,1:numSam_train);
% sample.test = sample.all(:,:,:,numSam_train+1:end);
% col = 11;row = 1;
% sam_instance = reshape(sample.all(:,:,:,1:numSam_train),[4096 size(sample.all,2) 11*numSam_train]);
% lab_instan = repmat(lab_trn,[1 11])';
% lab_instan = lab_instan(:);
%tst_instance = reshape(sample.all(:,:,:,numSam_train+1:end),[4096 size(sample.all,2) 11*numSam_test]);
%------ cluster patches of training sample in iLayer ------%
%samCluster_all = [reshape(sample.train,dPCA,55*numSam_train)]';
% Lab123 = repmat(lab_trn,1,11)';
% samImgLabel = Lab123(:); %每个特征对应的标签
aaa = [];
for iSeg = seg_c%[1 6 7 8 9]%1:se
iSeg
numSam = numSam_train + numSam_test;
samCluster = reshape(data_train(:,iSeg,:,:),[4096 11*numSam_train]);
lab_instan = repmat(lab_trn,[1 11])';
lab_instan = lab_instan(:);
samCluster = samCluster'; %[N1*11 4096]
if ~strcmp(clusterfun,'no')&&(numCluster_list(iSeg)~= 0)
numCluster = numCluster_list(iSeg);
if strcmp(clusterfun,'dplim')
res_cluster = func_dpCluster(samCluster,numCluster);
end
if strcmp(clusterfun,'kmlim')
new_path2 = 'I:\wangyancheng\code\ntu\cluster_data';
if ~exist(new_path2)
mkdir(new_path2); % 若不存在，在当前目录中产生一个子目录
end
filename = ['k' num2str(numCluster) '_s_test' num2str(iSeg) '.mat'];
if ~exist([new_path2 '\' filename])
res_cluster = func_kmCluster1(samCluster,numCluster,1000,1);
save([new_path2 '\' filename],'res_cluster');
else
load([new_path2 '\' filename])
end
end
cluster_Label = res_cluster.labels;
%% find the cluster center sample
if 0
dSam=samCluster;%[numSam,~]
cluster_center =res_cluster.centersFea;%[numCenter,~]
samDist = func_distCenter(dSam,cluster_center); %NxC
for i = 1:30
cluster_indx = find(samDist(:,i) == min(samDist(:,i)), 1 );
file_idx = floor(cluster_indx/11);
view_idx = mod(cluster_indx,11)
fileNames_Tr(file_idx,:)
disp('------------------');
end
end
ratioEachCl= zeros(30,60);
ratioEachClass = zeros(30,60);
%% Global
for i = 1:numCluster
numInCluster(i) = length(find(cluster_Label == i)); %各个簇所含特征数量
for j = 1:numClass
idxClassJ = find(lab_instan == j); %标签j在所有特征中的位置
clusterLabelForClassJ = cluster_Label(idxClassJ); %标签j下的特征所属簇 找到标签j的簇label
numEachClInClass(i,j) = length(find(clusterLabelForClassJ == i)); %第i个簇中第j标签的特征数量
ratioEachClass(i,j) = numEachClInClass(i,j)/length(idxClassJ);
end
ratioEachCl(i,:) = numEachClInClass(i,:) / numInCluster(i); % 每个簇在不同类别下的分布
ratioEachCl(i,ratioEachCl(i,:)==0 ) = 0.0001;
globalEntropy(i) = 1 + sum(ratioEachCl(i,:).*log(ratioEachCl(i,:)))/log(numClass);
end
%% Local
HI_score = []
IJ = [];
for i = 1:numClass
for j = i+1:numClass
HI_score = [HI_score, 1-sum(min(ratioEachClass(:,i),ratioEachCl(:,j)))];
IJ = [IJ;i,j];
end
end
id = HI_score>0.95;
a_l = IJ(id,:);
a_l = a_l(:);
B = unique(a_l);
c = zeros(length(B),1);
[a,id] = sort(c);
Dis_F = 1
if length(id)>10
numLocal = 10;
elseif length(id)>0
numLocal = length(id);
else
Dis_F = 0
end
id = id(end-int8(numLocal)+s1:end);
localClass = B(id);
if Dis_F
for k = 1:numCluster
ratioLocalCl = numEachClInClass(k,localClass) / numInCluster(k);
ratioLocalCl(ratioLocalCl==0) = 0.0001;
ratioLocalCl = ratioLocalCl/sum(ratioLocalCl);
localEntropy(k) = 1 + sum(ratioLocalCl.*log2(ratioLocalCl))/log2(length(localClass));
end
else
localEntropy = 0;
end
useLocal=0;
if useLocal ==1
beta = 0.3;
else
beta=0;
end
gamma = 0.016;
idx_discard = [];
%score(i,:) = globalEntropy.*(beta+localEntropy(i,:));
% score(i,:) = globalEntropy;
% gamma = 0.004 ;% (1-class_acc(i))/15;
% gamma = 0.025 - find(idx_rank==i)/400;
% idxlocal = find(localEntropy(i,:)<gamma);
% idx_c = idxlocal(globalEntropy(idxlocal)<0.1);
score = (1-beta)*globalEntropy + beta*localEntropy;
% [scorePattern idxscore] = sort(Score(i,:));
% idx_c = find(score(i,:) <gamma); %找到得分小于阈值的pattern
[a,id] = sort(score);
idx_c = id(1);
for k = 1:length(idx_c)
idx = find(cluster_Label==idx_c(k));
idx_discard = [idx_discard; idx]; %找到sam_instance中最终要丢弃的instance
end
aaa(iSeg) = length(idx_discard)
zz = randperm(length(idx_discard));
%idx_discard = idx_discard(zz(1:length(idx_discard)/2))';
%%%%%%%%%%%%%% used to select sample for GMM
%------ select patches ------%
samGmm = samCluster;
samGmm(idx_discard,:) = [];
clear samCluster;
%------ learning gmm ------%
disp('Computing the gmm model......');
[gmmComp.mean,gmmComp.covariances,gmmComp.priors] = vl_gmm(samGmm',numGmm);%[4096 N1*11]
if ~isdir('gmmComp\')
mkdir('gmmComp\');
end
save(['gmmComp\' 'resGmm_' num2str(iSeg) '.mat'],'gmmComp');
disp('gmmComp have been saved!');
clear samGmm
%------ test instance select ------%
%%%%%%%%%%%%%%% used to select sample for FV
%----- FV coding -----%
b = ones(numSam_train*11,1);
b(idx_discard) = 0;
b = reshape(b,11,numSam_train)';
for iSam = 1:numSam_train
dSam = squeeze(data_train(:,iSeg,:,iSam))';
dSam(b(iSam,:)==0,:)=[];
if isFVNorm
feaFV = vl_fisher(dSam',gmmComp.mean,gmmComp.covariances,gmmComp.priors,'Normalized');
else
feaFV = vl_fisher(dSam',gmmComp.mean,gmmComp.covariances,gmmComp.priors,'Improved');
end
%-- store the fisher vector---%
if ~isdir('finalFeaTR\')
mkdir('finalFeaTR\');
end
feaName = ['FV_feaTr' num2str(iSam) '_seg' num2str(iSeg)];
save(['finalFeaTR\',feaName,'.mat'],'feaFV');
disp(['Got ',num2str(iSam),'/',num2str(numSam),'--',num2str(iSeg),num2str(se) ' samples' ' final feature data .....']);
end
for iSam= 1:numSam_test
if ~isdir('finalFeaTs\')
mkdir('finalFeaTs\');
end
idxCenterChoose =idx_c;
fvSam = squeeze(data_test(:,iSeg,:,iSam))';
samDist = func_distCenter(fvSam,res_cluster.centersFea);
for i = 1:11
samLabel(i) = find(samDist(i,:) == min(samDist(i,:)), 1 );
end
idxSelect = find(samLabel == idxCenterChoose);
if length(idxSelect) >4
iidx = randperm(length(idxSelect));
nnumSel = 4;
idxSelect = idxSelect(iidx(1:nnumSel));
end
fvSam(idxSelect,:) = [];
feaFV = vl_fisher(fvSam',gmmComp.mean,gmmComp.covariances,gmmComp.priors,'Improved');
feaName = ['FV_feaTs' num2str(iSam) '_seg' num2str(iSeg)];
save(['finalFeaTs\',feaName,'.mat'],'feaFV');
end
else
% samGmm = samCluster;%[N1*11 4096]
% clear samCluster;
%------ learning gmm ------%
%[4096 N1*11]
disp('Computing the cluster model......');
%centers = vl_kmeans(dataLearn, numClusters);
centers = vl_kmeans(single(samCluster'),numC);
kdtree = vl_kdtreebuild(centers) ;
if ~isdir('gmmComp\')
mkdir('gmmComp\');
end
%%save(['gmmComp\' 'resGmm_' num2str(iSeg) '.mat'],'gmmComp');
%disp('gmmComp have been saved!');
disp('gmm have been kearned!');
%----- FV coding -----%
for iSam = 1:numSam_train
dSam = single(squeeze(data_train(:,iSeg,:,iSam)));%4096*11
[index, dist] = vl_kdtreequery(kdtree, centers, dSam) ;
[~, y] = size(dSam);%11
assignments = zeros(numC, y);
assignments(sub2ind(size(assignments), index, 1:length(index))) = 1;
assignments = single(assignments);
feaVLAD = vl_vlad(dSam, centers, assignments);
%-- store the VlAD---%
if ~isdir('final_vladTR\')
mkdir('final_vladTR\');
end
feaName = ['FV_feaTr' num2str(iSam) '_seg' num2str(iSeg)];
save(['final_vladTR\',feaName,'.mat'],'feaVLAD');
disp(['Got ',num2str(iSam),'/',num2str(numSam_train),'--',num2str(iSeg),'/',num2str(se) ' samples' ' final feature data .....']);
end
for iSam= 1:numSam_test
if ~isdir('final_vladTs\')
mkdir('final_vladTs\');
end
dSam = single(squeeze(data_test(:,iSeg,:,iSam)));%4096*11
[index, dist] = vl_kdtreequery(kdtree, centers, dSam) ;
[~, y] = size(dSam);%11
assignments = zeros(numC, y);
assignments(sub2ind(size(assignments), index, 1:length(index))) = 1;
assignments = single(assignments);
feaVLAD = vl_vlad(dSam, centers, assignments);
feaName = ['FV_feaTs' num2str(iSam) '_seg' num2str(iSeg)];
save(['final_vladTs\',feaName,'.mat'],'feaVLAD');
disp(['Got ',num2str(iSam),'/',num2str(numSam_test),'--',num2str(iSeg),'/',num2str(se) ' samples' ' final feature data .....']);
end
end
clear sub_sample
end
clear feaName feaFV dsamgmm gmmComp samgmm b
%----- training SVM -------%
samTrain = [];
disp('Porcessing all training samples......');
training_label_vector = double(lab_trn);
for i = 1:numSam_train
i
tempSamFea = [];
for iSeg = seg_c
for iBlock = 1:numBlock
feaName = ['FV_feaTr' num2str(i) '_seg' num2str(iSeg)];
load(['final_vladTR\',feaName,'.mat'],'feaVLAD');
tempSamFea = [tempSamFea; feaVLAD];
end
end
tempSamFea = tempSamFea';
samTrain = [samTrain;tempSamFea];
end
training_instance_sparse = sparse(double(samTrain));
disp('Training svm model......');
svmModel = train(training_label_vector, training_instance_sparse);
samTest = [];
disp('Processing all testing samples......');
testing_label_vector = lab_tst;
for i = 1:numSam_test
tempSamFea = [];
for iSeg = seg_c%1:se
for iBlock = 1:numBlock
feaName = ['FV_feaTs' num2str(i) '_seg' num2str(iSeg)];
load(['final_vladTs\',feaName,'.mat'],'feaVLAD');
tempSamFea = [tempSamFea; feaVLAD];
end
end
tempSamFea = tempSamFea';
samTest = [samTest;tempSamFea];
end
testing_instance_sparse = sparse(double(samTest));
disp('Testing......');
[predicted_label,accuracy,decision_values] = ...
predict(testing_label_vector, testing_instance_sparse, svmModel);
% predict_label= predicted_label;
%
% for i = 1:numClass
% num_in_class(i) = length(find(lab_tst==i));
%
% end
% name_class = importdata('classInd.txt');
% for ci = 1:numClass
% for cj = 1:numClass
% c_end = sum(num_in_class(1:ci));
% c_start = c_end - num_in_class(ci)+1;
% confusion_matrix(ci,cj)=length(find(predict_label(c_start:c_end)==cj))/num_in_class(ci);
%
% end
% end
%
% draw_cm(confusion_matrix,name_class,numClass);