-
Notifications
You must be signed in to change notification settings - Fork 5
/
trainer.cpp
487 lines (421 loc) · 13.6 KB
/
trainer.cpp
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
#define _CRT_SECURE_NO_WARNINGS
#include <iostream>
#include "opencvHeader.h"
#include <direct.h>
#include <omp.h>
#include "Algorithm.hpp"
#include "SLTP.h"
#include "LTDP.h"
#include "HDD.h"
#include "LDPK.h"
#include "ELDP.h"
#include <vector>
#include <sstream>
#include "Utils.h"
using namespace std;
using namespace cv;
using namespace cv::ml;
string hardpath;
string modelpath;
bool getSamples(const string& path, vector<string>& filenames)
{
ifstream sap(path.c_str());
if (!sap.is_open())
{
cerr << "sample open failed" << endl;
return false;
}
string filename;
while (getline(sap, filename) && !filename.empty())
{
filenames.push_back(filename);
}
return true;
}
void Train(DetectionAlgorithm* detectors[], string detectornames[], const string& pospath, const string& negpath, const string& negoripath,
const string& hardfilepath, const string& modelpaths)
{
_mkdir(hardfilepath.c_str());
_mkdir(modelpaths.c_str());
vector<string> posfiles;
vector<string> negfiles;
vector<string> negorifiles;
Utils::findallfiles(pospath, posfiles,"png");
Utils::findallfiles(negpath, negfiles,"png");
Utils::findallfiles(negoripath, negorifiles, "png");
//string detectornames[4] = { "SLTP","HDD","LTDP","HONV" };
//#pragma omp parallel for
for (int j = 2; j < 3; ++j)
{
Ptr<SVM> mysvm = SVM::create();
mysvm->setKernel(SVM::LINEAR);
mysvm->setType(SVM::C_SVC);
mysvm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER, 100, 1e-6));
Mat FeatureMat, LabelMat;
int featurelen = detectors[j]->getFeatureLen();
FeatureMat = Mat::zeros(posfiles.size() + negfiles.size(), featurelen, CV_32FC1);
LabelMat = Mat::zeros(posfiles.size() + negfiles.size(), 1, CV_32S);
cout << "读取正样本" << endl;
cout << "正样本数量:" << posfiles.size() << endl;
#pragma omp parallel for
for (int i = 0; i < posfiles.size(); ++i)
{
//cout << i << endl;
string path = pospath + posfiles[i];
//cout << posfiles[i] << endl;
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path, IMREAD_ANYDEPTH);
if (sample.rows != 128 || sample.cols != 64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
//filterimg=sample;
detectors[j]->compute(sample, description);
float* ptr = FeatureMat.ptr<float>(i);
memcpy(ptr, &description[0], sizeof(float)*featurelen);
LabelMat.at<int>(i, 0) = 1;
}
cout << "正样本计算完毕" << endl;
cout << "读取负样本" << endl;
cout << "负样本数量:" << negfiles.size() << endl;
#pragma omp parallel for
for (int i = 0; i < negfiles.size(); ++i)
{
//cout << i << endl;
//cout << negfiles[i] << endl;
string path = negpath + negfiles[i];
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path, IMREAD_ANYDEPTH);
if (sample.rows != 128 || sample.cols != 64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
detectors[j]->compute(sample, description);
float* ptr = FeatureMat.ptr<float>(i + posfiles.size());
memcpy(ptr, &description[0], sizeof(float)*featurelen);
LabelMat.at<int>(i + posfiles.size(), 0) = -1;
}
//第一轮训练
cout << "开始训练" << endl;
Ptr<TrainData> tData = TrainData::create(FeatureMat, ROW_SAMPLE, LabelMat);
//mysvm->trainAuto(tData);
mysvm->train(tData);
mysvm->save(modelpaths + detectornames[j] + "_"+traintype+"svm1.xml");
cout << "第一轮训练完毕,boosstrap" << endl;
//计算platt_scaling参数
Mat preLabel(LabelMat.size(), CV_32SC1);
preLabel.setTo(-1);
cout << "platt_scaling" << endl;
for (int i = 0; i < posfiles.size();++i)
{
string path = pospath + posfiles[i];
//cout << posfiles[i] << endl;
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path, IMREAD_ANYDEPTH);
if (sample.rows != 128 || sample.cols != 64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
//filterimg=sample;
vector<Rect> founds;
vector<double> weights;
detectors[j]->detectMultiScale(sample, founds, weights);
if (!founds.empty())
{
preLabel.at<int>(i, 0) = 1;
}
}
for (int i = 0; i < negfiles.size(); ++i)
{
string path = negpath + negfiles[i];
//cout << posfiles[i] << endl;
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path, IMREAD_ANYDEPTH);
if (sample.rows != 128 || sample.cols != 64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
Mat sample;
if (pfunc == NULL)
{
sample = sample;
}
else
pfunc(sample, sample);
//filterimg=sample;
vector<Rect> founds;
vector<double> weights;
detectors[j]->detectMultiScale(sample, founds, weights);
if (!founds.empty())
{
preLabel.at<int>(i+posfiles.size(), 0) = 1;
}
}
//bootstrap
for (int h = 0; h < 1; ++h)
{
char hardName[256];
sprintf(hardName, "%s%d\\", (hardfilepath+detectornames[j]+"\\"+traintype+"\\").c_str(), h);
_mkdir((hardfilepath + detectornames[j]).c_str());
_mkdir((hardfilepath + detectornames[j] + "\\" + traintype + "\\").c_str());
_mkdir(hardName);
string hardtemppath(hardName);
int cursamplesize = FeatureMat.rows;
detectors[j]->setSvmDetector(mysvm);
for (int i = 0; i < negorifiles.size(); ++i)
{
//cout << i << endl;
string path = negoripath + negorifiles[i];
Mat sample = imread(path, IMREAD_ANYDEPTH);
Mat sample;
if (pfunc == NULL)
{
sample = sample;
}
else
pfunc(sample, sample);
vector<Rect> found;
vector<double> weights;
detectors[j]->detectMultiScale(sample, found, weights);
for (int j = 0; j < found.size(); ++j)
{
//检测出来的很多矩形框都超出了图像边界,将这些矩形框都强制规范在图像边界内部
Rect r = found[j];
if (r.x < 0)
r.x = 0;
if (r.y < 0)
r.y = 0;
if (r.x + r.width > sample.cols)
r.width = sample.cols - r.x;
if (r.y + r.height > sample.rows)
r.height = sample.rows - r.y;
//将矩形框保存为图片,就是Hard Example
Mat hardExampleImg = sample(r);//从原图上截取矩形框大小的图片
char saveName[256];//裁剪出来的负样本图片文件名
string hardsavepath = hardtemppath + negorifiles[i];
hardsavepath.erase(hardsavepath.end() - 4, hardsavepath.end());
resize(hardExampleImg, hardExampleImg, Size(64, 128), INTER_NEAREST);//将剪裁出来的图片缩放为64*128大小
sprintf(saveName, "%s-%02d.png", hardsavepath.c_str(), j);//生成hard example图片的文件名
imwrite(saveName, hardExampleImg);//保存文件
}
found.clear();
sample.release();
}
vector<string> hardfiles;
Utils::findallfiles(hardtemppath, hardfiles, "png");
cout << "错误分类数: " << hardfiles.size() << endl;
if (hardfiles.size()<10)
{
break;
}
FeatureMat.resize(FeatureMat.rows + hardfiles.size());
LabelMat.resize(LabelMat.rows + hardfiles.size());
#pragma omp parallel for
for (int i = 0; i < hardfiles.size(); ++i)
{
string path = hardtemppath + hardfiles[i];
//cout << hardfiles[i] << endl;
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path, IMREAD_ANYDEPTH);
if (sample.rows != 128 || sample.cols != 64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
//filterimg = sample;
Mat sample;
if (pfunc == NULL)
{
sample = sample;
}
else
pfunc(sample, sample);
detectors[j]->compute(sample, description);
float* ptr = FeatureMat.ptr<float>(i + cursamplesize);
memcpy(ptr, &description[0], sizeof(float)*featurelen);
LabelMat.at<int>(i + cursamplesize, 0) = -1;
}
//train again
cout << "再次训练: " << h << endl;
tData = TrainData::create(FeatureMat, ROW_SAMPLE, LabelMat);
//mysvm->trainAuto(tData);
mysvm->train(tData);
cout << "训练完毕" << endl;
char svmname[256];
sprintf(svmname, (modelpaths + detectornames[j] + "_" + traintype+ "svm%d.xml").c_str(), h + 2);
mysvm->save(svmname);
}
}
}
//训练单个分类器
void Train(DetectionAlgorithm* detector,const string& pospath, const vector<string>& posfiles,
const string& negpath, const vector<string>& negfiles, const string& negoripath, const vector<string>& negorifiles)
{
Ptr<SVM> mysvm = SVM::create();
mysvm->setKernel(SVM::LINEAR);
mysvm->setType(SVM::C_SVC);
mysvm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER, 100, 1e-6));
Mat FeatureMat;
Mat LabelMat;
int featurelen = detector->getFeatureLen();
FeatureMat = Mat::zeros(posfiles.size() + negfiles.size(), featurelen, CV_32FC1);
LabelMat = Mat::zeros(posfiles.size() + negfiles.size(), 1, CV_32S);
//读取正样本
cout << "读取正样本" << endl;
cout << "正样本数量:" << posfiles.size() << endl;
#pragma omp parallel for default(none) shared(detector,FeatureMat,featurelen,LabelMat,pospath,posfiles)
for (int i = 0; i < posfiles.size();++i)
{
string path = pospath + posfiles[i];
//cout << posfiles[i] << endl;
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path,IMREAD_ANYDEPTH);
if (sample.rows!= 128 ||sample.cols!=64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
//filterimg=sample;
detector->compute(sample, description);
float* ptr = FeatureMat.ptr<float>(i);
memcpy(ptr, &description[0], sizeof(float)*featurelen);
LabelMat.at<int>(i, 0) = 1;
}
cout << "正样本计算完毕" << endl;
cout << "读取负样本" << endl;
cout << "负样本数量:" << negfiles.size() << endl;
#pragma omp parallel for default(none) shared(detector,FeatureMat,featurelen,LabelMat,negpath,negfiles,posfiles)
for (int i = 0; i < negfiles.size(); ++i)
{
//cout << negfiles[i] << endl;
string path = negpath + negfiles[i];
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path, IMREAD_ANYDEPTH);
if (sample.rows != 128 || sample.cols != 64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
//filterimg=sample;
detector->compute(sample, description);
float* ptr = FeatureMat.ptr<float>(i+posfiles.size());
memcpy(ptr, &description[0], sizeof(float)*featurelen);
LabelMat.at<int>(i+posfiles.size(), 0) = -1;
}
//设置训练数据
cout << "开始训练" << endl;
Ptr<TrainData> tData = TrainData::create(FeatureMat, ROW_SAMPLE, LabelMat);
//mysvm->trainAuto(tData);
mysvm->train(tData);
cout << "第一轮训练完毕" << endl;
//在原始负样本上测试
mysvm->save((modelpath + "sltpsvmfm1.xml").c_str());
for (int h = 0; h < 2;++h)
{
char hardName[256];
sprintf(hardName, "%s%d\\", hardpath.c_str(), h);
_mkdir(hardName);
string hardtemppath(hardName);
detector->setSvmDetector(mysvm);
int cursamplesize = FeatureMat.rows;
//cout << cursamplesize << endl;
#pragma omp parallel for
for (int i = 0; i < negorifiles.size(); ++i)
{
//cout << i << endl;
string path = negoripath + negorifiles[i];
Mat sample = imread(path, IMREAD_ANYDEPTH);
Mat filterimg;
preProcessing::pixelFilter(sample, filterimg);
//filterimg = sample;
vector<Rect> found;
vector<double> weights;
vector<Point> ts;
//honv.detect(sample, ts, weights);
detector->detectMultiScale(filterimg, found, weights);
/*if (found.size() > 0)
cout << found.size() << endl;*/
for (int j = 0; j < found.size(); ++j)
{
//检测出来的很多矩形框都超出了图像边界,将这些矩形框都强制规范在图像边界内部
Rect r = found[j];
if (r.x < 0)
r.x = 0;
if (r.y < 0)
r.y = 0;
if (r.x + r.width > sample.cols)
r.width = sample.cols - r.x;
if (r.y + r.height > sample.rows)
r.height = sample.rows - r.y;
//将矩形框保存为图片,就是Hard Example
Mat hardExampleImg = sample(r);//从原图上截取矩形框大小的图片
char saveName[256];//裁剪出来的负样本图片文件名
string hardsavepath = hardtemppath + negorifiles[i];
hardsavepath.erase(hardsavepath.end() - 4, hardsavepath.end());
resize(hardExampleImg, hardExampleImg, Size(64, 128), INTER_NEAREST);//将剪裁出来的图片缩放为64*128大小
sprintf(saveName, "%s-%02d.png", hardsavepath.c_str(), j);//生成hard example图片的文件名
imwrite(saveName, hardExampleImg);//保存文件
}
found.clear();
sample.release();
}
vector<string> hardfiles;
Utils::findallfiles(hardtemppath, hardfiles, "png");
cout << "错误分类数: " << hardfiles.size() << endl;
if (hardfiles.size()<10)
{
break;
}
FeatureMat.resize(FeatureMat.rows + hardfiles.size());
LabelMat.resize(LabelMat.rows + hardfiles.size());
#pragma omp parallel for
for (int i = 0; i < hardfiles.size();++i)
{
string path = hardtemppath + hardfiles[i];
//cout << hardfiles[i] << endl;
vector<float> description;
description.reserve(featurelen);
Mat sample = imread(path, IMREAD_ANYDEPTH);
if (sample.rows != 128 || sample.cols != 64)
{
resize(sample, sample, Size(64, 128), 0.0, 0.0, INTER_NEAREST);
}
detector->compute(sample, description);
float* ptr = FeatureMat.ptr<float>(i+ cursamplesize);
memcpy(ptr, &description[0], sizeof(float)*featurelen);
LabelMat.at<int>(i+ cursamplesize, 0) = -1;
}
//train again
cout << "再次训练: " <<h<< endl;
tData = TrainData::create(FeatureMat, ROW_SAMPLE, LabelMat);
//mysvm->trainAuto(tData);
mysvm->train(tData);
cout << "训练完毕" << endl;
char svmname[256];
sprintf(svmname, (modelpath+"sltpsvmfm1%d.xml").c_str(), h+2);
mysvm->save(svmname);
}
//提取hardfiles的特征
//mysvm->save("sltpsvm2.xml");
cout << "训练完毕" << endl;
}
int main()
{
string negpath = "";
string pospath = "";
string negoripath = "";
hardpath = "";
modelpath = "";
string featurepath = "";
DetectionAlgorithm* detectors[10];
//Train here
for (int i = 0; i < 10;++i)
{
delete detectors[i];
}
system("pause");
}