Merge pull request #3238 from vpisarev:bfmatcher_fix

pull/3242/head
Vadim Pisarevsky 10 years ago
commit 197b2e75e1
  1. 8
      modules/core/src/stat.cpp
  2. 15
      modules/features2d/src/matchers.cpp
  3. 91
      modules/features2d/test/test_matchers_algorithmic.cpp

@ -2463,14 +2463,14 @@ struct BatchDistInvoker : public ParallelLoopBody
}
void cv::batchDistance( InputArray _src1, InputArray _src2,
OutputArray _dist, int dtype, OutputArray _nidx,
int normType, int K, InputArray _mask,
int update, bool crosscheck )
OutputArray _dist, int dtype, OutputArray _nidx,
int normType, int K, InputArray _mask,
int update, bool crosscheck )
{
Mat src1 = _src1.getMat(), src2 = _src2.getMat(), mask = _mask.getMat();
int type = src1.type();
CV_Assert( type == src2.type() && src1.cols == src2.cols &&
(type == CV_32F || type == CV_8U));
(type == CV_32F || type == CV_8U));
CV_Assert( _nidx.needed() == (K > 0) );
if( dtype == -1 )

@ -352,18 +352,27 @@ void BFMatcher::knnMatchImpl( const Mat& queryDescriptors, vector<vector<DMatch>
matches.reserve(queryDescriptors.rows);
Mat dist, nidx;
int iIdx, imgCount = (int)trainDescCollection.size(), update = 0;
int dtype = normType == NORM_HAMMING || normType == NORM_HAMMING2 ||
(normType == NORM_L1 && queryDescriptors.type() == CV_8U) ? CV_32S : CV_32F;
int maxRows = 0;
CV_Assert( (int64)imgCount*IMGIDX_ONE < INT_MAX );
for( iIdx = 0; iIdx < imgCount; iIdx++ )
maxRows = std::max(maxRows, trainDescCollection[iIdx].rows);
int m = queryDescriptors.rows;
Mat dist(m, knn, dtype), nidx(m, knn, CV_32S);
dist = Scalar::all(dtype == CV_32S ? (double)INT_MAX : (double)FLT_MAX);
nidx = Scalar::all(-1);
for( iIdx = 0; iIdx < imgCount; iIdx++ )
{
CV_Assert( trainDescCollection[iIdx].rows < IMGIDX_ONE );
batchDistance(queryDescriptors, trainDescCollection[iIdx], dist, dtype, nidx,
int n = std::min(knn, trainDescCollection[iIdx].rows);
Mat dist_i = dist.colRange(0, n), nidx_i = nidx.colRange(0, n);
batchDistance(queryDescriptors, trainDescCollection[iIdx], dist_i, dtype, nidx_i,
normType, knn, masks.empty() ? Mat() : masks[iIdx], update, crossCheck);
update += IMGIDX_ONE;
}

@ -57,6 +57,9 @@ public:
CV_DescriptorMatcherTest( const string& _name, const Ptr<DescriptorMatcher>& _dmatcher, float _badPart ) :
badPart(_badPart), name(_name), dmatcher(_dmatcher)
{}
static void generateData( Mat& query, Mat& train );
protected:
static const int dim = 500;
static const int queryDescCount = 300; // must be even number because we split train data in some cases in two
@ -64,7 +67,6 @@ protected:
const float badPart;
virtual void run( int );
void generateData( Mat& query, Mat& train );
void emptyDataTest();
void matchTest( const Mat& query, const Mat& train );
@ -526,6 +528,81 @@ void CV_DescriptorMatcherTest::run( int )
radiusMatchTest( query, train );
}
// bug #3172: test that knnMatch() can handle images with fewer than knn keypoints
class CV_DescriptorMatcherLowKeypointTest : public cvtest::BaseTest
{
public:
CV_DescriptorMatcherLowKeypointTest( const string& _name, const Ptr<DescriptorMatcher>& _dmatcher ) :
name(_name), dmatcher(_dmatcher)
{}
protected:
virtual void run(int);
void knnMatchTest( const Mat& query, const Mat& train );
private:
string name;
Ptr<DescriptorMatcher> dmatcher;
};
void CV_DescriptorMatcherLowKeypointTest::knnMatchTest( const Mat& query, const Mat& train )
{
const int knn = 6;
const int queryDescCount = query.rows;
vector<vector<DMatch> > matches;
// three train images, the third one with only one keypoint
dmatcher->add( vector<Mat>(1,train.rowRange(0, train.rows/2)) );
dmatcher->add( vector<Mat>(1,train.rowRange(train.rows/2, train.rows-1)) );
dmatcher->add( vector<Mat>(1,train.rowRange(train.rows-1, train.rows)) );
const int trainImgCount = (int)dmatcher->getTrainDescriptors().size();
dmatcher->knnMatch( query, matches, knn, std::vector<Mat>(), true );
if( matches.empty() )
{
ts->printf(cvtest::TS::LOG, "No matches while testing knnMatch() function (3).\n");
ts->set_failed_test_info( cvtest::TS::FAIL_INVALID_OUTPUT );
}
else
{
int badImgIdxCount = 0, badQueryIdxCount = 0, badTrainIdxCount = 0;
for( size_t i = 0; i < matches.size(); i++ )
{
for( size_t j = 0; j < matches[i].size(); j++ )
{
const DMatch& match = matches[i][j];
if( match.imgIdx < 0 || match.imgIdx >= trainImgCount )
{
++badImgIdxCount;
}
if( match.queryIdx < 0 || match.queryIdx >= queryDescCount )
{
++badQueryIdxCount;
}
if( match.trainIdx < 0 )
{
++badTrainIdxCount;
}
}
}
if( badImgIdxCount > 0 || badQueryIdxCount > 0 || badTrainIdxCount > 0 )
{
ts->printf( cvtest::TS::LOG, "%d/%d/%d - wrong image/query/train indices while testing knnMatch() function (3).\n",
badImgIdxCount, badQueryIdxCount, badTrainIdxCount );
ts->set_failed_test_info( cvtest::TS::FAIL_INVALID_OUTPUT );
}
}
}
void CV_DescriptorMatcherLowKeypointTest::run( int )
{
Mat query, train;
CV_DescriptorMatcherTest::generateData( query, train );
knnMatchTest( query, train );
}
/****************************************************************************************\
* Tests registrations *
\****************************************************************************************/
@ -541,3 +618,15 @@ TEST( Features2d_DescriptorMatcher_FlannBased, regression )
CV_DescriptorMatcherTest test( "descriptor-matcher-flann-based", Algorithm::create<DescriptorMatcher>("DescriptorMatcher.FlannBasedMatcher"), 0.04f );
test.safe_run();
}
TEST( Features2d_DescriptorMatcher_LowKeypoint_BruteForce, regression )
{
CV_DescriptorMatcherLowKeypointTest test( "descriptor-matcher-low-keypoint-brute-force", Algorithm::create<DescriptorMatcher>("DescriptorMatcher.BFMatcher") );
test.safe_run();
}
TEST(Features2d_DescriptorMatcher_LowKeypoint_FlannBased, regression)
{
CV_DescriptorMatcherLowKeypointTest test( "descriptor-matcher-low-keypoint-flann-based", Algorithm::create<DescriptorMatcher>("DescriptorMatcher.FlannBasedMatcher") );
test.safe_run();
}

Loading…
Cancel
Save