opencv/samples/cpp/train_HOG.cpp

393 lines
13 KiB
C++
Raw Normal View History

2016-06-26 08:42:10 +08:00
#include "opencv2/imgproc.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/ml.hpp"
#include "opencv2/objdetect.hpp"
#include <iostream>
#include <time.h>
using namespace cv;
using namespace cv::ml;
using namespace std;
2017-11-19 06:41:24 +08:00
vector< float > get_svm_detector( const Ptr< SVM >& svm );
2016-06-26 08:42:10 +08:00
void convert_to_ml( const std::vector< Mat > & train_samples, Mat& trainData );
void load_images( const String & dirname, vector< Mat > & img_lst, bool showImages );
2013-11-12 02:29:39 +08:00
void sample_neg( const vector< Mat > & full_neg_lst, vector< Mat > & neg_lst, const Size & size );
2017-11-19 06:41:24 +08:00
void computeHOGs( const Size wsize, const vector< Mat > & img_lst, vector< Mat > & gradient_lst, bool use_flip );
void test_trained_detector( String obj_det_filename, String test_dir, String videofilename );
2017-11-19 06:41:24 +08:00
vector< float > get_svm_detector( const Ptr< SVM >& svm )
{
// get the support vectors
Mat sv = svm->getSupportVectors();
const int sv_total = sv.rows;
// get the decision function
Mat alpha, svidx;
2016-06-26 08:42:10 +08:00
double rho = svm->getDecisionFunction( 0, alpha, svidx );
CV_Assert( alpha.total() == 1 && svidx.total() == 1 && sv_total == 1 );
CV_Assert( (alpha.type() == CV_64F && alpha.at<double>(0) == 1.) ||
(alpha.type() == CV_32F && alpha.at<float>(0) == 1.f) );
CV_Assert( sv.type() == CV_32F );
2017-11-19 06:41:24 +08:00
vector< float > hog_detector( sv.cols + 1 );
2016-06-26 08:42:10 +08:00
memcpy( &hog_detector[0], sv.ptr(), sv.cols*sizeof( hog_detector[0] ) );
hog_detector[sv.cols] = (float)-rho;
2017-11-19 06:41:24 +08:00
return hog_detector;
}
/*
* Convert training/testing set to be used by OpenCV Machine Learning algorithms.
* TrainData is a matrix of size (#samples x max(#cols,#rows) per samples), in 32FC1.
* Transposition of samples are made if needed.
*/
2016-06-26 08:42:10 +08:00
void convert_to_ml( const vector< Mat > & train_samples, Mat& trainData )
{
//--Convert data
const int rows = (int)train_samples.size();
const int cols = (int)std::max( train_samples[0].cols, train_samples[0].rows );
2016-06-26 08:42:10 +08:00
Mat tmp( 1, cols, CV_32FC1 ); //< used for transposition if needed
trainData = Mat( rows, cols, CV_32FC1 );
for( size_t i = 0 ; i < train_samples.size(); ++i )
{
2016-06-26 08:42:10 +08:00
CV_Assert( train_samples[i].cols == 1 || train_samples[i].rows == 1 );
if( train_samples[i].cols == 1 )
{
2016-06-26 08:42:10 +08:00
transpose( train_samples[i], tmp );
tmp.copyTo( trainData.row( (int)i ) );
}
2016-06-26 08:42:10 +08:00
else if( train_samples[i].rows == 1 )
{
2016-06-26 08:42:10 +08:00
train_samples[i].copyTo( trainData.row( (int)i ) );
}
}
}
2016-06-26 08:42:10 +08:00
void load_images( const String & dirname, vector< Mat > & img_lst, bool showImages = false )
{
2016-06-26 08:42:10 +08:00
vector< String > files;
glob( dirname, files );
2016-06-26 08:42:10 +08:00
for ( size_t i = 0; i < files.size(); ++i )
{
2016-06-26 08:42:10 +08:00
Mat img = imread( files[i] ); // load the image
if ( img.empty() ) // invalid image, skip it.
{
cout << files[i] << " is invalid!" << endl;
continue;
}
2016-06-26 08:42:10 +08:00
if ( showImages )
2013-11-24 22:32:52 +08:00
{
2016-06-26 08:42:10 +08:00
imshow( "image", img );
waitKey( 1 );
2013-11-24 22:32:52 +08:00
}
2016-06-26 08:42:10 +08:00
img_lst.push_back( img );
}
}
void sample_neg( const vector< Mat > & full_neg_lst, vector< Mat > & neg_lst, const Size & size )
{
Rect box;
box.width = size.width;
box.height = size.height;
const int size_x = box.width;
const int size_y = box.height;
2013-11-24 22:32:52 +08:00
srand( (unsigned int)time( NULL ) );
2016-06-26 08:42:10 +08:00
for ( size_t i = 0; i < full_neg_lst.size(); i++ )
2017-11-19 06:41:24 +08:00
if ( full_neg_lst[i].cols >= box.width && full_neg_lst[i].rows >= box.height )
{
box.x = rand() % ( full_neg_lst[i].cols - size_x );
box.y = rand() % ( full_neg_lst[i].rows - size_y );
Mat roi = full_neg_lst[i]( box );
neg_lst.push_back( roi.clone() );
}
}
2017-11-19 06:41:24 +08:00
void computeHOGs( const Size wsize, const vector< Mat > & img_lst, vector< Mat > & gradient_lst, bool use_flip )
2013-11-24 22:32:52 +08:00
{
2016-06-26 08:42:10 +08:00
HOGDescriptor hog;
hog.winSize = wsize;
Mat gray;
vector< float > descriptors;
2017-11-19 06:41:24 +08:00
for( size_t i = 0 ; i < img_lst.size(); i++ )
{
2017-11-19 06:41:24 +08:00
if ( img_lst[i].cols >= wsize.width && img_lst[i].rows >= wsize.height )
{
Rect r = Rect(( img_lst[i].cols - wsize.width ) / 2,
( img_lst[i].rows - wsize.height ) / 2,
wsize.width,
wsize.height);
cvtColor( img_lst[i](r), gray, COLOR_BGR2GRAY );
hog.compute( gray, descriptors, Size( 8, 8 ), Size( 0, 0 ) );
gradient_lst.push_back( Mat( descriptors ).clone() );
if ( use_flip )
{
flip( gray, gray, 1 );
hog.compute( gray, descriptors, Size( 8, 8 ), Size( 0, 0 ) );
gradient_lst.push_back( Mat( descriptors ).clone() );
}
}
2016-06-26 08:42:10 +08:00
}
}
2017-11-19 06:41:24 +08:00
void test_trained_detector( String obj_det_filename, String test_dir, String videofilename )
2016-06-26 08:42:10 +08:00
{
cout << "Testing trained detector..." << endl;
HOGDescriptor hog;
hog.load( obj_det_filename );
2016-06-26 08:42:10 +08:00
vector< String > files;
glob( test_dir, files );
2016-06-26 08:42:10 +08:00
int delay = 0;
VideoCapture cap;
2016-06-26 08:42:10 +08:00
if ( videofilename != "" )
{
2017-11-19 06:41:24 +08:00
if ( videofilename.size() == 1 && isdigit( videofilename[0] ) )
cap.open( videofilename[0] - '0' );
else
cap.open( videofilename );
2016-06-26 08:42:10 +08:00
}
2016-06-26 08:42:10 +08:00
obj_det_filename = "testing " + obj_det_filename;
namedWindow( obj_det_filename, WINDOW_NORMAL );
2016-06-26 08:42:10 +08:00
for( size_t i=0;; i++ )
{
Mat img;
2016-06-26 08:42:10 +08:00
if ( cap.isOpened() )
{
cap >> img;
delay = 1;
}
else if( i < files.size() )
{
img = imread( files[i] );
}
2016-06-26 08:42:10 +08:00
if ( img.empty() )
{
2017-11-19 06:41:24 +08:00
return;
2016-06-26 08:42:10 +08:00
}
2016-06-26 08:42:10 +08:00
vector< Rect > detections;
vector< double > foundWeights;
2016-06-26 08:42:10 +08:00
hog.detectMultiScale( img, detections, foundWeights );
for ( size_t j = 0; j < detections.size(); j++ )
{
2016-06-26 08:42:10 +08:00
Scalar color = Scalar( 0, foundWeights[j] * foundWeights[j] * 200, 0 );
rectangle( img, detections[j], color, img.cols / 400 + 1 );
}
2016-06-26 08:42:10 +08:00
imshow( obj_det_filename, img );
2017-11-19 06:41:24 +08:00
if( waitKey( delay ) == 27 )
2016-06-26 08:42:10 +08:00
{
2017-11-19 06:41:24 +08:00
return;
}
}
2016-06-26 08:42:10 +08:00
}
2016-06-26 08:42:10 +08:00
int main( int argc, char** argv )
{
const char* keys =
{
2016-06-26 08:42:10 +08:00
"{help h| | show help message}"
"{pd | | path of directory contains possitive images}"
"{nd | | path of directory contains negative images}"
"{td | | path of directory contains test images}"
"{tv | | test video file name}"
"{dw | | width of the detector}"
"{dh | | height of the detector}"
2017-11-19 06:41:24 +08:00
"{f |false| indicates if the program will generate and use mirrored samples or not}"
2016-06-26 08:42:10 +08:00
"{d |false| train twice}"
"{t |false| test a trained detector}"
"{v |false| visualize training steps}"
"{fn |my_detector.yml| file name of trained SVM}"
};
CommandLineParser parser( argc, argv, keys );
if ( parser.has( "help" ) )
{
parser.printMessage();
exit( 0 );
}
2016-06-26 08:42:10 +08:00
String pos_dir = parser.get< String >( "pd" );
String neg_dir = parser.get< String >( "nd" );
String test_dir = parser.get< String >( "td" );
String obj_det_filename = parser.get< String >( "fn" );
String videofilename = parser.get< String >( "tv" );
int detector_width = parser.get< int >( "dw" );
int detector_height = parser.get< int >( "dh" );
bool test_detector = parser.get< bool >( "t" );
bool train_twice = parser.get< bool >( "d" );
bool visualization = parser.get< bool >( "v" );
2017-11-19 06:41:24 +08:00
bool flip_samples = parser.get< bool >( "f" );
2016-06-26 08:42:10 +08:00
if ( test_detector )
{
test_trained_detector( obj_det_filename, test_dir, videofilename );
exit( 0 );
}
2016-06-26 08:42:10 +08:00
if( pos_dir.empty() || neg_dir.empty() )
{
parser.printMessage();
cout << "Wrong number of parameters.\n\n"
2017-11-19 06:41:24 +08:00
<< "Example command line:\n" << argv[0] << " -dw=64 -dh=128 -pd=/INRIAPerson/96X160H96/Train/pos -nd=/INRIAPerson/neg -td=/INRIAPerson/Test/pos -fn=HOGpedestrian64x128.xml -d\n"
<< "\nExample command line for testing trained detector:\n" << argv[0] << " -t -fn=HOGpedestrian64x128.xml -td=/INRIAPerson/Test/pos";
2016-06-26 08:42:10 +08:00
exit( 1 );
}
2016-06-26 08:42:10 +08:00
vector< Mat > pos_lst, full_neg_lst, neg_lst, gradient_lst;
vector< int > labels;
2016-06-26 08:42:10 +08:00
clog << "Positive images are being loaded..." ;
load_images( pos_dir, pos_lst, visualization );
if ( pos_lst.size() > 0 )
{
clog << "...[done]" << endl;
}
else
{
clog << "no image in " << pos_dir <<endl;
return 1;
}
2016-06-26 08:42:10 +08:00
Size pos_image_size = pos_lst[0].size();
2016-06-26 08:42:10 +08:00
if ( detector_width && detector_height )
{
2016-06-26 08:42:10 +08:00
pos_image_size = Size( detector_width, detector_height );
}
2017-11-19 06:41:24 +08:00
else
{
for ( size_t i = 0; i < pos_lst.size(); ++i )
{
if( pos_lst[i].size() != pos_image_size )
{
cout << "All positive images should be same size!" << endl;
exit( 1 );
}
}
pos_image_size = pos_image_size / 8 * 8;
}
2016-06-26 08:42:10 +08:00
clog << "Negative images are being loaded...";
load_images( neg_dir, full_neg_lst, false );
sample_neg( full_neg_lst, neg_lst, pos_image_size );
clog << "...[done]" << endl;
clog << "Histogram of Gradients are being calculated for positive images...";
2017-11-19 06:41:24 +08:00
computeHOGs( pos_image_size, pos_lst, gradient_lst, flip_samples );
size_t positive_count = gradient_lst.size();
labels.assign( positive_count, +1 );
clog << "...[done] ( positive count : " << positive_count << " )" << endl;
2016-06-26 08:42:10 +08:00
clog << "Histogram of Gradients are being calculated for negative images...";
2017-11-19 06:41:24 +08:00
computeHOGs( pos_image_size, neg_lst, gradient_lst, flip_samples );
size_t negative_count = gradient_lst.size() - positive_count;
labels.insert( labels.end(), negative_count, -1 );
CV_Assert( positive_count < labels.size() );
clog << "...[done] ( negative count : " << negative_count << " )" << endl;
Mat train_data;
convert_to_ml( gradient_lst, train_data );
2016-06-26 08:42:10 +08:00
clog << "Training SVM...";
Ptr< SVM > svm = SVM::create();
/* Default values to train SVM */
2016-06-26 08:42:10 +08:00
svm->setCoef0( 0.0 );
svm->setDegree( 3 );
svm->setTermCriteria( TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 1000, 1e-3 ) );
2016-06-26 08:42:10 +08:00
svm->setGamma( 0 );
svm->setKernel( SVM::LINEAR );
svm->setNu( 0.5 );
svm->setP( 0.1 ); // for EPSILON_SVR, epsilon in loss function?
svm->setC( 0.01 ); // From paper, soft classifier
svm->setType( SVM::EPS_SVR ); // C_SVC; // EPSILON_SVR; // may be also NU_SVR; // do regression task
2017-11-19 06:41:24 +08:00
svm->train( train_data, ROW_SAMPLE, labels );
clog << "...[done]" << endl;
2016-06-26 08:42:10 +08:00
if ( train_twice )
{
2016-06-26 08:42:10 +08:00
clog << "Testing trained detector on negative images. This may take a few minutes...";
HOGDescriptor my_hog;
my_hog.winSize = pos_image_size;
2016-06-26 08:42:10 +08:00
// Set the trained svm to my_hog
2017-11-19 06:41:24 +08:00
my_hog.setSVMDetector( get_svm_detector( svm ) );
2016-06-26 08:42:10 +08:00
vector< Rect > detections;
vector< double > foundWeights;
2016-06-26 08:42:10 +08:00
for ( size_t i = 0; i < full_neg_lst.size(); i++ )
{
2017-11-19 06:41:24 +08:00
if ( full_neg_lst[i].cols >= pos_image_size.width && full_neg_lst[i].rows >= pos_image_size.height )
my_hog.detectMultiScale( full_neg_lst[i], detections, foundWeights );
else
detections.clear();
2016-06-26 08:42:10 +08:00
for ( size_t j = 0; j < detections.size(); j++ )
{
Mat detection = full_neg_lst[i]( detections[j] ).clone();
resize( detection, detection, pos_image_size, 0, 0, INTER_LINEAR_EXACT);
2016-06-26 08:42:10 +08:00
neg_lst.push_back( detection );
}
2017-11-19 06:41:24 +08:00
2016-06-26 08:42:10 +08:00
if ( visualization )
{
for ( size_t j = 0; j < detections.size(); j++ )
{
rectangle( full_neg_lst[i], detections[j], Scalar( 0, 255, 0 ), 2 );
}
imshow( "testing trained detector on negative images", full_neg_lst[i] );
waitKey( 5 );
}
}
clog << "...[done]" << endl;
2016-06-26 08:42:10 +08:00
gradient_lst.clear();
clog << "Histogram of Gradients are being calculated for positive images...";
2017-11-19 06:41:24 +08:00
computeHOGs( pos_image_size, pos_lst, gradient_lst, flip_samples );
positive_count = gradient_lst.size();
clog << "...[done] ( positive count : " << positive_count << " )" << endl;
2016-06-26 08:42:10 +08:00
clog << "Histogram of Gradients are being calculated for negative images...";
2017-11-19 06:41:24 +08:00
computeHOGs( pos_image_size, neg_lst, gradient_lst, flip_samples );
negative_count = gradient_lst.size() - positive_count;
clog << "...[done] ( negative count : " << negative_count << " )" << endl;
labels.clear();
labels.assign(positive_count, +1);
labels.insert(labels.end(), negative_count, -1);
2016-06-26 08:42:10 +08:00
clog << "Training SVM again...";
convert_to_ml( gradient_lst, train_data );
2017-11-19 06:41:24 +08:00
svm->train( train_data, ROW_SAMPLE, labels );
2016-06-26 08:42:10 +08:00
clog << "...[done]" << endl;
}
2016-06-26 08:42:10 +08:00
HOGDescriptor hog;
hog.winSize = pos_image_size;
2017-11-19 06:41:24 +08:00
hog.setSVMDetector( get_svm_detector( svm ) );
2016-06-26 08:42:10 +08:00
hog.save( obj_det_filename );
2016-06-26 08:42:10 +08:00
test_trained_detector( obj_det_filename, test_dir, videofilename );
return 0;
}