generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Normalize line endings and whitespace

Browse Source
This commit is contained in:
OpenCV Buildbot
2012-10-17 11:12:04 +04:00
committed by Andrey Kamaev
parent 0442bca235
commit 81f826db2b
1511 changed files with 258678 additions and 258624 deletions
Download Patch File Download Diff File Expand all files Collapse all files

106
samples/cpp/detector_descriptor_matcher_evaluation.cpp
View File

@@ -59,40 +59,40 @@ using namespace cv;
The algorithm:
for each tested combination of detector+descriptor+matcher:
create detector, descriptor and matcher,
load their params if they are there, otherwise use the default ones and save them
for each dataset:
load reference image
detect keypoints in it, compute descriptors
for each transformed image:
load the image
load the transformation matrix
detect keypoints in it too, compute descriptors
find matches
transform keypoints from the first image using the ground-truth matrix
compute the number of matched keypoints, i.e. for each pair (i,j) found by a matcher compare
j-th keypoint from the second image with the transformed i-th keypoint. If they are close, +1.
so, we have:
N - number of keypoints in the first image that are also visible
(after transformation) on the second image
N1 - number of keypoints in the first image that have been matched.
n - number of the correct matches found by the matcher
n/N1 - precision
n/N - recall (?)
we store (N, n/N1, n/N) (where N is stored primarily for tuning the detector's thresholds,
in order to semi-equalize their keypoints counts)
*/
typedef Vec3f TVec; // (N, n/N1, n/N) - see above
@@ -138,7 +138,7 @@ static void transformKeypoints( const vector<KeyPoint>& kp,
size_t i, n = kp.size();
contours.resize(n);
vector<Point> temp;
for( i = 0; i < n; i++ )
{
ellipse2Poly(Point2f(kp[i].pt.x*scale, kp[i].pt.y*scale),
@@ -157,7 +157,7 @@ static TVec proccessMatches( Size imgsize,
double overlapThreshold )
{
const double visibilityThreshold = 0.6;
// 1. [preprocessing] find bounding rect for each element of kp1t_contours and kp_contours.
// 2. [cross-check] for each DMatch (iK, i1)
// update best_match[i1] using DMatch::distance.
@@ -167,18 +167,18 @@ static TVec proccessMatches( Size imgsize,
// if best_match[i1] is initialized, increment N1
// if kp_contours[best_match[i1]] and kp1t_contours[i1] overlap by overlapThreshold*100%,
// increment n. Use bounding rects to speedup this step
int i, size1 = (int)kp1t_contours.size(), size = (int)kp_contours.size(), msize = (int)matches.size();
vector<DMatch> best_match(size1);
vector<Rect> rects1(size1), rects(size);
// proprocess
for( i = 0; i < size1; i++ )
rects1[i] = boundingRect(kp1t_contours[i]);
for( i = 0; i < size; i++ )
rects[i] = boundingRect(kp_contours[i]);
// cross-check
for( i = 0; i < msize; i++ )
{
@@ -188,35 +188,35 @@ static TVec proccessMatches( Size imgsize,
if( best_match[i1].trainIdx < 0 || best_match[i1].distance > m.distance )
best_match[i1] = m;
}
int N = 0, N1 = 0, n = 0;
// overlapping
for( i = 0; i < size1; i++ )
{
int i1 = i, iK = best_match[i].queryIdx;
if( iK >= 0 )
N1++;
Rect r = rects1[i] & Rect(0, 0, imgsize.width, imgsize.height);
if( r.area() < visibilityThreshold*rects1[i].area() )
continue;
N++;
if( iK < 0 || (rects1[i1] & rects[iK]).area() == 0 )
continue;
double n_area = intersectConvexConvex(kp1t_contours[i1], kp_contours[iK], noArray(), true);
if( n_area == 0 )
continue;
double area1 = contourArea(kp1t_contours[i1], false);
double area = contourArea(kp_contours[iK], false);
double ratio = n_area/(area1 + area - n_area);
n += ratio >= overlapThreshold;
}
return TVec((float)N, (float)n/std::max(N1, 1), (float)n/std::max(N, 1));
}
@@ -228,7 +228,7 @@ static void saveResults(const string& dir, const string& name, const string& dsn
string fname2 = format("%s%s_%s_recall.csv", dir.c_str(), name.c_str(), dsname.c_str());
FILE* f1 = fopen(fname1.c_str(), "wt");
FILE* f2 = fopen(fname2.c_str(), "wt");
for( size_t i = 0; i < results.size(); i++ )
{
fprintf(f1, "%d, %.1f\n", xvals[i], results[i][1]*100);
@@ -251,31 +251,31 @@ int main(int argc, char** argv)
//"SURF_BF", "SURF", "SURF", "BruteForce",
0
};
static const char* datasets[] =
{
"bark", "bikes", "boat", "graf", "leuven", "trees", "ubc", "wall", 0
};
static const int imgXVals[] = { 2, 3, 4, 5, 6 }; // if scale, blur or light changes
static const int viewpointXVals[] = { 20, 30, 40, 50, 60 }; // if viewpoint changes
static const int jpegXVals[] = { 60, 80, 90, 95, 98 }; // if jpeg compression
const double overlapThreshold = 0.6;
vector<vector<vector<TVec> > > results; // indexed as results[ddm][dataset][testcase]
string dataset_dir = string(getenv("OPENCV_TEST_DATA_PATH")) +
"/cv/detectors_descriptors_evaluation/images_datasets";
string dir=argc > 1 ? argv[1] : ".";
if( dir[dir.size()-1] != '\\' && dir[dir.size()-1] != '/' )
dir += "/";
int result = system(("mkdir " + dir).c_str());
CV_Assert(result == 0);
for( int i = 0; ddms[i*4] != 0; i++ )
{
const char* name = ddms[i*4];
@@ -283,40 +283,40 @@ int main(int argc, char** argv)
const char* descriptor_name = ddms[i*4+2];
const char* matcher_name = ddms[i*4+3];
string params_filename = dir + string(name) + "_params.yml";
cout << "Testing " << name << endl;
Ptr<FeatureDetector> detector = FeatureDetector::create(detector_name);
Ptr<DescriptorExtractor> descriptor = DescriptorExtractor::create(descriptor_name);
Ptr<DescriptorMatcher> matcher = DescriptorMatcher::create(matcher_name);
saveloadDDM( params_filename, detector, descriptor, matcher );
results.push_back(vector<vector<TVec> >());
for( int j = 0; datasets[j] != 0; j++ )
{
const char* dsname = datasets[j];
cout << "\ton " << dsname << " ";
cout.flush();
const int* xvals = strcmp(dsname, "ubc") == 0 ? jpegXVals :
strcmp(dsname, "graf") == 0 || strcmp(dsname, "wall") == 0 ? viewpointXVals : imgXVals;
vector<KeyPoint> kp1, kp;
vector<DMatch> matches;
vector<vector<Point2f> > kp1t_contours, kp_contours;
Mat desc1, desc;
Mat img1 = imread(format("%s/%s/img1.png", dataset_dir.c_str(), dsname), 0);
CV_Assert( !img1.empty() );
detector->detect(img1, kp1);
descriptor->compute(img1, kp1, desc1);
results[i].push_back(vector<TVec>());
for( int k = 2; ; k++ )
{
cout << ".";
@@ -324,20 +324,20 @@ int main(int argc, char** argv)
Mat imgK = imread(format("%s/%s/img%d.png", dataset_dir.c_str(), dsname, k), 0);
if( imgK.empty() )
break;
detector->detect(imgK, kp);
descriptor->compute(imgK, kp, desc);
matcher->match( desc, desc1, matches );
Mat H = loadMat(format("%s/%s/H1to%dp.xml", dataset_dir.c_str(), dsname, k));
transformKeypoints( kp1, kp1t_contours, H );
transformKeypoints( kp, kp_contours, Mat::eye(3, 3, CV_64F));
TVec r = proccessMatches( imgK.size(), matches, kp1t_contours, kp_contours, overlapThreshold );
results[i][j].push_back(r);
}
saveResults(dir, name, dsname, results[i][j], xvals);
cout << endl;
}