Skip to content

O
opencv
Commit 88bd1f1d authored by Ilya Lysenkov's avatar Ilya Lysenkov
Browse files
Options

Added plot data generation for visual descriptors comparison in the evaluation framework

parent f6f634ba
Hide whitespace changes
Inline Side-by-side
Showing with 241 additions and 19 deletions
modules/features2d/include/opencv2/features2d/features2d.hpp
View file @ 88bd1f1d
......@@ -1503,6 +1503,25 @@ struct CV_EXPORTS L2
}
};
/****************************************************************************************\
* DescriptorMatching *
\****************************************************************************************/
/*
* Struct for matching: match index and distance between descriptors
*/
struct DescriptorMatching
{
int index;
float distance;
//less is better
bool operator<( const DescriptorMatching &m) const
{
return distance < m.distance;
}
};
/****************************************************************************************\
* DescriptorMatcher *
\****************************************************************************************/
......@@ -1545,6 +1564,28 @@ public:
void match( const Mat& query, const Mat& mask,
vector<int>& matches ) const;
/*
* Find the best match for each descriptor from a query set
*
* query The query set of descriptors
* matchings Matchings of the closest matches from the training set
*/
void match( const Mat& query, vector<DescriptorMatching>& matchings ) const;
/*
* Find the best matches between two descriptor sets, with constraints
* on which pairs of descriptors can be matched.
*
* The mask describes which descriptors can be matched. descriptors_1[i]
* can be matched with descriptors_2[j] only if mask.at<char>(i,j) is non-zero.
*
* query The query set of descriptors
* mask Mask specifying permissible matches.
* matchings Matchings of the closest matches from the training set
*/
void match( const Mat& query, const Mat& mask,
vector<DescriptorMatching>& matchings ) const;
/*
* Find the best keypoint matches for small view changes.
*
......@@ -1574,6 +1615,13 @@ protected:
virtual void matchImpl( const Mat& descriptors_1, const Mat& descriptors_2,
const Mat& mask, vector<int>& matches ) const = 0;
/*
* Find matches; match() calls this. Must be implemented by the subclass.
* The mask may be empty.
*/
virtual void matchImpl( const Mat& descriptors_1, const Mat& descriptors_2,
const Mat& mask, vector<DescriptorMatching>& matches ) const = 0;
static bool possibleMatch( const Mat& mask, int index_1, int index_2 )
{
return mask.empty() || mask.at<char>(index_1, index_2);
......@@ -1609,6 +1657,18 @@ inline void DescriptorMatcher::match( const Mat& query, const Mat& mask,
matchImpl( query, train, mask, matches );
}
inline void DescriptorMatcher::match( const Mat& query, vector<DescriptorMatching>& matches ) const
{
matchImpl( query, train, Mat(), matches );
}
inline void DescriptorMatcher::match( const Mat& query, const Mat& mask,
vector<DescriptorMatching>& matches ) const
{
matchImpl( query, train, mask, matches );
}
inline void DescriptorMatcher::clear()
{
train.release();
......@@ -1633,12 +1693,28 @@ protected:
virtual void matchImpl( const Mat& descriptors_1, const Mat& descriptors_2,
const Mat& mask, vector<int>& matches ) const;
virtual void matchImpl( const Mat& descriptors_1, const Mat& descriptors_2,
const Mat& mask, vector<DescriptorMatching>& matches ) const;
Distance distance;
};
template<class Distance>
void BruteForceMatcher<Distance>::matchImpl( const Mat& descriptors_1, const Mat& descriptors_2,
const Mat& mask, vector<int>& matches ) const
{
vector<DescriptorMatching> matchings;
matchImpl( descriptors_1, descriptors_2, mask, matchings);
matches.resize( matchings.size() );
for( size_t i=0;i<matchings.size();i++)
{
matches[i] = matchings[i].index;
}
}
template<class Distance>
void BruteForceMatcher<Distance>::matchImpl( const Mat& descriptors_1, const Mat& descriptors_2,
const Mat& mask, vector<DescriptorMatching>& matches ) const
{
typedef typename Distance::ValueType ValueType;
typedef typename Distance::ResultType DistanceType;
......@@ -1650,8 +1726,7 @@ void BruteForceMatcher<Distance>::matchImpl( const Mat& descriptors_1, const Mat
assert( DataType<ValueType>::type == descriptors_2.type() || descriptors_2.empty() );
int dimension = descriptors_1.cols;
matches.clear();
matches.reserve(descriptors_1.rows);
matches.resize(descriptors_1.rows);
for( int i = 0; i < descriptors_1.rows; i++ )
{
......@@ -1674,7 +1749,12 @@ void BruteForceMatcher<Distance>::matchImpl( const Mat& descriptors_1, const Mat
}
if( matchIndex != -1 )
matches.push_back( matchIndex );
{
DescriptorMatching matching;
matching.index = matchIndex;
matching.distance = matchDistance;
matches[i] = matching;
}
}
}
......@@ -1742,6 +1822,12 @@ public:
// indices A vector to be filled with keypoint class indices
virtual void match( const Mat& image, vector<KeyPoint>& points, vector<int>& indices ) = 0;
// Matches test keypoints to the training set
// image The source image
// points Test keypoints from the source image
// matchings A vector to be filled with keypoint matchings
virtual void match( const Mat& image, vector<KeyPoint>& points, vector<DescriptorMatching>& matchings ) {};
// Clears keypoints storing in collection
virtual void clear();
......@@ -1816,6 +1902,8 @@ public:
// loaded with DescriptorOneWay::Initialize, kd tree is used for finding minimum distances.
virtual void match( const Mat& image, vector<KeyPoint>& points, vector<int>& indices );
virtual void match( const Mat& image, vector<KeyPoint>& points, vector<DescriptorMatching>& matchings );
// Classify a set of keypoints. The same as match, but returns point classes rather than indices
virtual void classify( const Mat& image, vector<KeyPoint>& points );
......@@ -1944,6 +2032,8 @@ public:
virtual void match( const Mat& image, vector<KeyPoint>& keypoints, vector<int>& indices );
virtual void match( const Mat& image, vector<KeyPoint>& points, vector<DescriptorMatching>& matchings );
virtual void classify( const Mat& image, vector<KeyPoint>& keypoints );
virtual void clear ();
......@@ -2000,6 +2090,14 @@ public:
matcher.match( descriptors, keypointIndices );
};
virtual void match( const Mat& image, vector<KeyPoint>& points, vector<DescriptorMatching>& matchings )
{
Mat descriptors;
extractor.compute( image, points, descriptors );
matcher.match( descriptors, matchings );
}
virtual void clear()
{
GenericDescriptorMatch::clear();
......
modules/features2d/src/descriptors.cpp
View file @ 88bd1f1d
......@@ -44,6 +44,8 @@
using namespace std;
using namespace cv;
//#define _KDTREE
/****************************************************************************************\
* DescriptorExtractor *
\****************************************************************************************/
......@@ -332,15 +334,27 @@ void OneWayDescriptorMatch::add( KeyPointCollection& keypoints )
void OneWayDescriptorMatch::match( const Mat& image, vector<KeyPoint>& points, vector<int>& indices)
{
vector<DescriptorMatching> matchings( points.size() );
indices.resize(points.size());
match( image, points, matchings );
for( size_t i = 0; i < points.size(); i++ )
indices[i] = matchings[i].index;
}
void OneWayDescriptorMatch::match( const Mat& image, vector<KeyPoint>& points, vector<DescriptorMatching>& matchings )
{
matchings.resize( points.size() );
IplImage _image = image;
for( size_t i = 0; i < points.size(); i++ )
{
int descIdx = -1;
int poseIdx = -1;
float distance;
base->FindDescriptor( &_image, points[i].pt, descIdx, poseIdx, distance );
indices[i] = descIdx;
DescriptorMatching matching;
matching.index = -1;
base->FindDescriptor( &_image, points[i].pt, matching.index, poseIdx, matching.distance );
matchings[i] = matching;
}
}
......@@ -631,6 +645,21 @@ void FernDescriptorMatch::match( const Mat& image, vector<KeyPoint>& keypoints,
}
}
void FernDescriptorMatch::match( const Mat& image, vector<KeyPoint>& keypoints, vector<DescriptorMatching>& matchings )
{
trainFernClassifier();
matchings.resize( keypoints.size() );
vector<float> signature( (size_t)classifier->getClassCount() );
for( size_t pi = 0; pi < keypoints.size(); pi++ )
{
calcBestProbAndMatchIdx( image, keypoints[pi].pt, matchings[pi].distance, matchings[pi].index, signature );
//matching[pi].distance is log of probability so we need to transform it
matchings[pi].distance = -matchings[pi].distance;
}
}
void FernDescriptorMatch::classify( const Mat& image, vector<KeyPoint>& keypoints )
{
trainFernClassifier();
......
tests/cv/src/adetectordescriptor_evaluation.cpp
View file @ 88bd1f1d
......@@ -549,9 +549,9 @@ inline float precision( int correctMatchCount, int falseMatchCount )
}
void evaluateDescriptors( const vector<EllipticKeyPoint>& keypoints1, const vector<EllipticKeyPoint>& keypoints2,
const vector<int>& matches1to2,
vector< pair<DescriptorMatching, int> >& matches1to2,
const Mat& img1, const Mat& img2, const Mat& H1to2,
int& correctMatchCount, int& falseMatchCount, int& correspondenceCount )
int &correctMatchCount, int &falseMatchCount, vector<int> &matchStatuses, int& correspondenceCount )
{
assert( !keypoints1.empty() && !keypoints2.empty() && !matches1to2.empty() );
assert( keypoints1.size() == matches1to2.size() );
......@@ -564,17 +564,27 @@ void evaluateDescriptors( const vector<EllipticKeyPoint>& keypoints1, const vect
repeatability, correspCount,
&thresholdedOverlapMask );
correspondenceCount = thresholdedOverlapMask.nzcount();
correctMatchCount = falseMatchCount = 0;
matchStatuses.resize( matches1to2.size() );
correctMatchCount = 0;
falseMatchCount = 0;
//the nearest descriptors should be examined first
std::sort( matches1to2.begin(), matches1to2.end() );
for( size_t i1 = 0; i1 < matches1to2.size(); i1++ )
{
int i2 = matches1to2[i1];
int i2 = matches1to2[i1].first.index;
if( i2 > 0 )
{
if( thresholdedOverlapMask(i1, i2) )
matchStatuses[i2] = thresholdedOverlapMask(matches1to2[i1].second, i2);
if( matchStatuses[i2] )
correctMatchCount++;
else
falseMatchCount++;
}
else
matchStatuses[i2] = -1;
}
}
......@@ -615,11 +625,16 @@ class BaseQualityTest : public CvTest
{
public:
BaseQualityTest( const char* _algName, const char* _testName, const char* _testFuncs ) :
CvTest( _testName, _testFuncs ), algName(_algName) {}
CvTest( _testName, _testFuncs ), algName(_algName)
{
//TODO: change this
isWriteGraphicsData = true;
}
protected:
virtual string getRunParamsFilename() const = 0;
virtual string getResultsFilename() const = 0;
virtual string getPlotPath() const = 0;
virtual void validQualityClear( int datasetIdx ) = 0;
virtual void calcQualityClear( int datasetIdx ) = 0;
......@@ -650,9 +665,11 @@ protected:
virtual void processResults();
virtual int processResults( int datasetIdx, int caseIdx ) = 0;
void writeAllPlotData() const;
virtual void writePlotData( const string &filename, int datasetIdx ) const {};
string algName;
bool isWriteParams, isWriteResults;
bool isWriteParams, isWriteResults, isWriteGraphicsData;
};
void BaseQualityTest::readAllDatasetsRunParams()
......@@ -811,6 +828,8 @@ void BaseQualityTest::processResults()
{
if( isWriteParams )
writeAllDatasetsRunParams();
if( isWriteGraphicsData )
writeAllPlotData();
int res = CvTS::OK;
if( isWriteResults )
......@@ -838,6 +857,18 @@ void BaseQualityTest::processResults()
ts->set_failed_test_info( res );
}
void BaseQualityTest::writeAllPlotData() const
{
for( int di = 0; di < DATASETS_COUNT; di++ )
{
stringstream stream;
stream << getPlotPath() << algName << "_" << DATASET_NAMES[di] << ".csv";
string filename;
stream >> filename;
writePlotData( filename, di );
}
}
void BaseQualityTest::run ( int )
{
readAlgorithm ();
......@@ -904,6 +935,7 @@ protected:
virtual string getRunParamsFilename() const;
virtual string getResultsFilename() const;
virtual string getPlotPath() const;
virtual void validQualityClear( int datasetIdx );
virtual void calcQualityClear( int datasetIdx );
......@@ -961,6 +993,11 @@ string DetectorQualityTest::getResultsFilename() const
return string(ts->get_data_path()) + DETECTORS_DIR + algName + RES_POSTFIX;
}
string DetectorQualityTest::getPlotPath() const
{
return string(ts->get_data_path()) + DETECTORS_DIR + "plots/";
}
void DetectorQualityTest::validQualityClear( int datasetIdx )
{
validQuality[datasetIdx].clear();
......@@ -1253,6 +1290,7 @@ public:
{
validQuality.resize(DATASETS_COUNT);
calcQuality.resize(DATASETS_COUNT);
calcDatasetQuality.resize(DATASETS_COUNT);
commRunParams.resize(DATASETS_COUNT);
commRunParamsDefault.projectKeypointsFrom1Image = true;
......@@ -1267,6 +1305,7 @@ protected:
virtual string getRunParamsFilename() const;
virtual string getResultsFilename() const;
virtual string getPlotPath() const;
virtual void validQualityClear( int datasetIdx );
virtual void calcQualityClear( int datasetIdx );
......@@ -1289,6 +1328,8 @@ protected:
virtual int processResults( int datasetIdx, int caseIdx );
virtual void writePlotData( const string &filename, int di ) const;
struct Quality
{
float recall;
......@@ -1296,6 +1337,7 @@ protected:
};
vector<vector<Quality> > validQuality;
vector<vector<Quality> > calcQuality;
vector<vector<Quality> > calcDatasetQuality;
struct CommonRunParams
{
......@@ -1322,6 +1364,11 @@ string DescriptorQualityTest::getResultsFilename() const
return string(ts->get_data_path()) + DESCRIPTORS_DIR + algName + RES_POSTFIX;
}
string DescriptorQualityTest::getPlotPath() const
{
return string(ts->get_data_path()) + DESCRIPTORS_DIR + "plots/";
}
void DescriptorQualityTest::validQualityClear( int datasetIdx )
{
validQuality[datasetIdx].clear();
......@@ -1408,6 +1455,16 @@ void DescriptorQualityTest::setDefaultDatasetRunParams( int datasetIdx )
commRunParams[datasetIdx].keypontsFilename = "surf_" + DATASET_NAMES[datasetIdx] + ".xml.gz";
}
void DescriptorQualityTest::writePlotData( const string &filename, int di ) const
{
FILE *file = fopen (filename.c_str(),"w");
size_t size = calcDatasetQuality[di].size();
for (size_t i=0;i<size;i++)
{
fprintf( file, "%f, %f\n", 1 - calcDatasetQuality[di][i].precision, calcDatasetQuality[di][i].recall);
}
fclose( file );
}
void DescriptorQualityTest::readAlgorithm( )
{
......@@ -1478,6 +1535,10 @@ void DescriptorQualityTest::runDatasetTest (const vector<Mat> &imgs, const vecto
transformToEllipticKeyPoints( keypoints1, ekeypoints1 );
int progressCount = DATASETS_COUNT*TEST_CASE_COUNT;
vector< pair<DescriptorMatching, int> > allMatchings;
vector<int> allMatchStatuses;
size_t matchingIndex = 0;
int allCorrespCount = 0;
for( int ci = 0; ci < TEST_CASE_COUNT; ci++ )
{
progress = update_progress( progress, di*TEST_CASE_COUNT + ci, progressCount, 0 );
......@@ -1494,16 +1555,50 @@ void DescriptorQualityTest::runDatasetTest (const vector<Mat> &imgs, const vecto
readKeypoints( keypontsFS, keypoints2, ci+1 );
transformToEllipticKeyPoints( keypoints2, ekeypoints2 );
descMatch->add( imgs[ci+1], keypoints2 );
vector<int> matches1to2;
descMatch->match( imgs[0], keypoints1, matches1to2 );
vector<DescriptorMatching> matchings1to2;
descMatch->match( imgs[0], keypoints1, matchings1to2 );
vector< pair<DescriptorMatching, int> > matchings (matchings1to2.size());
for( size_t i=0;i<matchings1to2.size();i++ )
matchings[i] = pair<DescriptorMatching, int>( matchings1to2[i], i);
// TODO if( commRunParams[di].matchFilter )
int correctMatchCount, falseMatchCount, correspCount;
evaluateDescriptors( ekeypoints1, ekeypoints2, matches1to2, imgs[0], imgs[ci+1], Hs[ci],
correctMatchCount, falseMatchCount, correspCount );
int correspCount;
int correctMatchCount = 0, falseMatchCount = 0;
vector<int> matchStatuses;
evaluateDescriptors( ekeypoints1, ekeypoints2, matchings, imgs[0], imgs[ci+1], Hs[ci],
correctMatchCount, falseMatchCount, matchStatuses, correspCount );
for( size_t i=0;i<matchings.size();i++ )
matchings[i].second += matchingIndex;
matchingIndex += matchings.size();
allCorrespCount += correspCount;
//TODO: use merge
std::copy( matchings.begin(), matchings.end(), std::back_inserter( allMatchings ) );
std::copy( matchStatuses.begin(), matchStatuses.end(), std::back_inserter( allMatchStatuses ) );
printf ("%d %d %d \n", correctMatchCount, falseMatchCount, correspCount );
calcQuality[di][ci].recall = recall( correctMatchCount, correspCount );
calcQuality[di][ci].precision = precision( correctMatchCount, falseMatchCount );
descMatch->clear ();
}
std::sort( allMatchings.begin(), allMatchings.end() );
calcDatasetQuality[di].resize( allMatchings.size() );
int correctMatchCount = 0, falseMatchCount = 0;
for( size_t i=0;i<allMatchings.size();i++)
{
if( allMatchStatuses[ allMatchings[i].second ] )
correctMatchCount++;
else
falseMatchCount++;
calcDatasetQuality[di][i].recall = recall( correctMatchCount, allCorrespCount );
calcDatasetQuality[di][i].precision = precision( correctMatchCount, falseMatchCount );
}
}
int DescriptorQualityTest::processResults( int datasetIdx, int caseIdx )
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment