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

Merge branch 'master' of https://github.com/Itseez/opencv

Browse Source 
Added python binding for createButton
This commit is contained in:
Matthew Skolaut
2016-06-20 16:24:23 -05:00
parent 9b959072a2 4597b94099
commit e8bfb48490
36 changed files with 1665 additions and 561 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
modules/calib3d/include/opencv2/calib3d.hpp
View File

@@ -767,6 +767,14 @@ k-th translation vector (see the next output parameter description) brings the c
from the model coordinate space (in which object points are specified) to the world coordinate
space, that is, a real position of the calibration pattern in the k-th pattern view (k=0.. *M* -1).
@param tvecs Output vector of translation vectors estimated for each pattern view.
@param stdDeviationsIntrinsics Output vector of standard deviations estimated for intrinsic parameters.
Order of deviations values:
\f$(f_x, f_y, c_x, c_y, k_1, k_2, p_1, p_2, k_3, k_4, k_5, k_6 , s_1, s_2, s_3,
s_4, \tau_x, \tau_y)\f$ If one of parameters is not estimated, it's deviation is equals to zero.
@param stdDeviationsExtrinsics Output vector of standard deviations estimated for extrinsic parameters.
Order of deviations values: \f$(R_1, T_1, \dotsc , R_M, T_M)\f$ where M is number of pattern views,
\f$R_i, T_i\f$ are concatenated 1x3 vectors.
@param perViewErrors Output vector of average re-projection errors estimated for each pattern view.
@param flags Different flags that may be zero or a combination of the following values:
- **CV_CALIB_USE_INTRINSIC_GUESS** cameraMatrix contains valid initial values of
fx, fy, cx, cy that are optimized further. Otherwise, (cx, cy) is initially set to the image
@@ -841,6 +849,24 @@ The function returns the final re-projection error.
@sa
findChessboardCorners, solvePnP, initCameraMatrix2D, stereoCalibrate, undistort
*/
CV_EXPORTS_AS(calibrateCameraExtended) double calibrateCamera( InputArrayOfArrays objectPoints,
InputArrayOfArrays imagePoints, Size imageSize,
InputOutputArray cameraMatrix, InputOutputArray distCoeffs,
OutputArrayOfArrays rvecs, OutputArrayOfArrays tvecs,
OutputArray stdDeviationsIntrinsics,
OutputArray stdDeviationsExtrinsics,
OutputArray perViewErrors,
int flags = 0, TermCriteria criteria = TermCriteria(
TermCriteria::COUNT + TermCriteria::EPS, 30, DBL_EPSILON) );
/** @overload double calibrateCamera( InputArrayOfArrays objectPoints,
InputArrayOfArrays imagePoints, Size imageSize,
InputOutputArray cameraMatrix, InputOutputArray distCoeffs,
OutputArrayOfArrays rvecs, OutputArrayOfArrays tvecs,
OutputArray stdDeviations, OutputArray perViewErrors,
int flags = 0, TermCriteria criteria = TermCriteria(
TermCriteria::COUNT + TermCriteria::EPS, 30, DBL_EPSILON) )
*/
CV_EXPORTS_W double calibrateCamera( InputArrayOfArrays objectPoints,
InputArrayOfArrays imagePoints, Size imageSize,
InputOutputArray cameraMatrix, InputOutputArray distCoeffs,

1
modules/calib3d/include/opencv2/calib3d/calib3d_c.h
View File

@@ -246,6 +246,7 @@ CVAPI(void) cvDrawChessboardCorners( CvArr* image, CvSize pattern_size,
#define CV_CALIB_TILTED_MODEL 262144
#define CV_CALIB_FIX_TAUX_TAUY 524288
#define CV_CALIB_NINTRINSIC 18
/* Finds intrinsic and extrinsic camera parameters
from a few views of known calibration pattern */

174
modules/calib3d/src/calibration.cpp
View File

@@ -1181,7 +1181,6 @@ CV_IMPL void cvFindExtrinsicCameraParams2( const CvMat* objectPoints,
cvConvert( &_t, tvec );
}
CV_IMPL void cvInitIntrinsicParams2D( const CvMat* objectPoints,
const CvMat* imagePoints, const CvMat* npoints,
CvSize imageSize, CvMat* cameraMatrix,
@@ -1270,15 +1269,37 @@ CV_IMPL void cvInitIntrinsicParams2D( const CvMat* objectPoints,
cvConvert( &_a, cameraMatrix );
}
static void subMatrix(const cv::Mat& src, cv::Mat& dst, const std::vector<uchar>& cols,
const std::vector<uchar>& rows) {
int nonzeros_cols = cv::countNonZero(cols);
cv::Mat tmp(src.rows, nonzeros_cols, CV_64FC1);
/* finds intrinsic and extrinsic camera parameters
from a few views of known calibration pattern */
CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
for (int i = 0, j = 0; i < (int)cols.size(); i++)
{
if (cols[i])
{
src.col(i).copyTo(tmp.col(j++));
}
}
int nonzeros_rows = cv::countNonZero(rows);
dst.create(nonzeros_rows, nonzeros_cols, CV_64FC1);
for (int i = 0, j = 0; i < (int)rows.size(); i++)
{
if (rows[i])
{
tmp.row(i).copyTo(dst.row(j++));
}
}
}
static double cvCalibrateCamera2Internal( const CvMat* objectPoints,
const CvMat* imagePoints, const CvMat* npoints,
CvSize imageSize, CvMat* cameraMatrix, CvMat* distCoeffs,
CvMat* rvecs, CvMat* tvecs, int flags, CvTermCriteria termCrit )
CvMat* rvecs, CvMat* tvecs, CvMat* stdDevs,
CvMat* perViewErrors, int flags, CvTermCriteria termCrit )
{
const int NINTRINSIC = 18;
const int NINTRINSIC = CV_CALIB_NINTRINSIC;
double reprojErr = 0;
Matx33d A;
@@ -1338,6 +1359,20 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
"1xn or nx1 array or 1-channel nx3 array, where n is the number of views" );
}
if( stdDevs )
{
cn = CV_MAT_CN(stdDevs->type);
if( !CV_IS_MAT(stdDevs) ||
(CV_MAT_DEPTH(stdDevs->type) != CV_32F && CV_MAT_DEPTH(stdDevs->type) != CV_64F) ||
((stdDevs->rows != (nimages*6 + NINTRINSIC) || stdDevs->cols*cn != 1) &&
(stdDevs->rows != 1 || stdDevs->cols != (nimages*6 + NINTRINSIC) || cn != 1)) )
#define STR__(x) #x
#define STR_(x) STR__(x)
CV_Error( CV_StsBadArg, "the output array of standard deviations vectors must be 1-channel "
"1x(n*6 + NINTRINSIC) or (n*6 + NINTRINSIC)x1 array, where n is the number of views,"
" NINTRINSIC = " STR_(CV_CALIB_NINTRINSIC));
}
if( (CV_MAT_TYPE(cameraMatrix->type) != CV_32FC1 &&
CV_MAT_TYPE(cameraMatrix->type) != CV_64FC1) ||
cameraMatrix->rows != 3 || cameraMatrix->cols != 3 )
@@ -1367,6 +1402,7 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
Mat matM( 1, total, CV_64FC3 );
Mat _m( 1, total, CV_64FC2 );
Mat allErrors(1, total, CV_64FC2);
if(CV_MAT_CN(objectPoints->type) == 3) {
cvarrToMat(objectPoints).convertTo(matM, CV_64F);
@@ -1518,6 +1554,7 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
double* _errNorm = 0;
bool proceed = solver.updateAlt( _param, _JtJ, _JtErr, _errNorm );
double *param = solver.param->data.db, *pparam = solver.prevParam->data.db;
bool calcJ = solver.state == CvLevMarq::CALC_J || (!proceed && stdDevs);
if( flags & CALIB_FIX_ASPECT_RATIO )
{
@@ -1528,8 +1565,10 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
A(0, 0) = param[0]; A(1, 1) = param[1]; A(0, 2) = param[2]; A(1, 2) = param[3];
std::copy(param + 4, param + 4 + 14, k);
if( !proceed )
if ( !proceed && !stdDevs && !perViewErrors )
break;
else if ( !proceed && stdDevs )
cvZero(_JtJ);
reprojErr = 0;
@@ -1543,6 +1582,7 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
CvMat _Mi(matM.colRange(pos, pos + ni));
CvMat _mi(_m.colRange(pos, pos + ni));
CvMat _me(allErrors.colRange(pos, pos + ni));
_Je.resize(ni*2); _Ji.resize(ni*2); _err.resize(ni*2);
CvMat _dpdr(_Je.colRange(0, 3));
@@ -1552,7 +1592,7 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
CvMat _dpdk(_Ji.colRange(4, NINTRINSIC));
CvMat _mp(_err.reshape(2, 1));
if( solver.state == CvLevMarq::CALC_J )
if( calcJ )
{
cvProjectPoints2( &_Mi, &_ri, &_ti, &matA, &_k, &_mp, &_dpdr, &_dpdt,
(flags & CALIB_FIX_FOCAL_LENGTH) ? 0 : &_dpdf,
@@ -1563,8 +1603,10 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
cvProjectPoints2( &_Mi, &_ri, &_ti, &matA, &_k, &_mp );
cvSub( &_mp, &_mi, &_mp );
if (perViewErrors || stdDevs)
cvCopy(&_mp, &_me);
if( solver.state == CvLevMarq::CALC_J )
if( calcJ )
{
Mat JtJ(cvarrToMat(_JtJ)), JtErr(cvarrToMat(_JtErr));
@@ -1581,15 +1623,52 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
}
if( _errNorm )
*_errNorm = reprojErr;
if( !proceed )
{
if( stdDevs )
{
Mat mask = cvarrToMat(solver.mask);
int nparams_nz = countNonZero(mask);
Mat JtJinv, JtJN;
JtJN.create(nparams_nz, nparams_nz, CV_64F);
subMatrix(cvarrToMat(_JtJ), JtJN, mask, mask);
completeSymm(JtJN, false);
cv::invert(JtJN, JtJinv, DECOMP_SVD);
//sigma2 is deviation of the noise
//see any papers about variance of the least squares estimator for
//detailed description of the variance estimation methods
double sigma2 = norm(allErrors, NORM_L2SQR) / (total - nparams_nz);
Mat stdDevsM = cvarrToMat(stdDevs);
int j = 0;
for ( int s = 0; s < nparams; s++ )
if( mask.data[s] )
{
stdDevsM.at<double>(s) = std::sqrt(JtJinv.at<double>(j,j) * sigma2);
j++;
}
else
stdDevsM.at<double>(s) = 0.;
}
break;
}
}
// 4. store the results
cvConvert( &matA, cameraMatrix );
cvConvert( &_k, distCoeffs );
for( i = 0; i < nimages; i++ )
for( i = 0, pos = 0; i < nimages; i++ )
{
CvMat src, dst;
if( perViewErrors )
{
ni = npoints->data.i[i*npstep];
perViewErrors->data.db[i] = std::sqrt(cv::norm(allErrors.colRange(pos, pos + ni),
NORM_L2SQR) / ni);
pos+=ni;
}
if( rvecs )
{
src = cvMat( 3, 1, CV_64F, solver.param->data.db + NINTRINSIC + i*6 );
@@ -1622,6 +1701,17 @@ CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
}
/* finds intrinsic and extrinsic camera parameters
from a few views of known calibration pattern */
CV_IMPL double cvCalibrateCamera2( const CvMat* objectPoints,
const CvMat* imagePoints, const CvMat* npoints,
CvSize imageSize, CvMat* cameraMatrix, CvMat* distCoeffs,
CvMat* rvecs, CvMat* tvecs, int flags, CvTermCriteria termCrit )
{
return cvCalibrateCamera2Internal(objectPoints, imagePoints, npoints, imageSize, cameraMatrix,
distCoeffs, rvecs, tvecs, NULL, NULL, flags, termCrit);
}
void cvCalibrationMatrixValues( const CvMat *calibMatr, CvSize imgSize,
double apertureWidth, double apertureHeight, double *fovx, double *fovy,
double *focalLength, CvPoint2D64f *principalPoint, double *pasp )
@@ -1772,7 +1862,7 @@ double cvStereoCalibrate( const CvMat* _objectPoints, const CvMat* _imagePoints1
if( !(flags & (CV_CALIB_FIX_INTRINSIC|CV_CALIB_USE_INTRINSIC_GUESS)))
{
cvCalibrateCamera2( objectPoints, imagePoints[k],
npoints, imageSize, &K[k], &Dist[k], 0, 0, flags );
npoints, imageSize, &K[k], &Dist[k], NULL, NULL, flags );
}
}
@@ -3091,7 +3181,6 @@ static void collectCalibrationData( InputArrayOfArrays objectPoints,
}
}
static Mat prepareCameraMatrix(Mat& cameraMatrix0, int rtype)
{
Mat cameraMatrix = Mat::eye(3, 3, rtype);
@@ -3287,10 +3376,23 @@ cv::Mat cv::initCameraMatrix2D( InputArrayOfArrays objectPoints,
}
double cv::calibrateCamera( InputArrayOfArrays _objectPoints,
InputArrayOfArrays _imagePoints,
Size imageSize, InputOutputArray _cameraMatrix, InputOutputArray _distCoeffs,
OutputArrayOfArrays _rvecs, OutputArrayOfArrays _tvecs, int flags, TermCriteria criteria )
{
return calibrateCamera(_objectPoints, _imagePoints, imageSize, _cameraMatrix, _distCoeffs,
_rvecs, _tvecs, noArray(), noArray(), noArray(), flags, criteria);
}
double cv::calibrateCamera(InputArrayOfArrays _objectPoints,
InputArrayOfArrays _imagePoints,
Size imageSize, InputOutputArray _cameraMatrix, InputOutputArray _distCoeffs,
OutputArrayOfArrays _rvecs, OutputArrayOfArrays _tvecs,
OutputArray stdDeviationsIntrinsics,
OutputArray stdDeviationsExtrinsics,
OutputArray _perViewErrors, int flags, TermCriteria criteria )
{
int rtype = CV_64F;
Mat cameraMatrix = _cameraMatrix.getMat();
@@ -3304,14 +3406,17 @@ double cv::calibrateCamera( InputArrayOfArrays _objectPoints,
int nimages = int(_objectPoints.total());
CV_Assert( nimages > 0 );
Mat objPt, imgPt, npoints, rvecM, tvecM;
Mat objPt, imgPt, npoints, rvecM, tvecM, stdDeviationsM, errorsM;
bool rvecs_needed = _rvecs.needed(), tvecs_needed = _tvecs.needed();
bool rvecs_needed = _rvecs.needed(), tvecs_needed = _tvecs.needed(),
stddev_needed = stdDeviationsIntrinsics.needed(), errors_needed = _perViewErrors.needed(),
stddev_ext_needed = stdDeviationsExtrinsics.needed();
bool rvecs_mat_vec = _rvecs.isMatVector();
bool tvecs_mat_vec = _tvecs.isMatVector();
if( rvecs_needed ) {
if( rvecs_needed )
{
_rvecs.create(nimages, 1, CV_64FC3);
if(rvecs_mat_vec)
@@ -3320,7 +3425,8 @@ double cv::calibrateCamera( InputArrayOfArrays _objectPoints,
rvecM = _rvecs.getMat();
}
if( tvecs_needed ) {
if( tvecs_needed )
{
_tvecs.create(nimages, 1, CV_64FC3);
if(tvecs_mat_vec)
@@ -3329,16 +3435,46 @@ double cv::calibrateCamera( InputArrayOfArrays _objectPoints,
tvecM = _tvecs.getMat();
}
if( stddev_needed || stddev_ext_needed )
{
stdDeviationsM.create(nimages*6 + CV_CALIB_NINTRINSIC, 1, CV_64F);
}
if( errors_needed )
{
_perViewErrors.create(nimages, 1, CV_64F);
errorsM = _perViewErrors.getMat();
}
collectCalibrationData( _objectPoints, _imagePoints, noArray(),
objPt, imgPt, 0, npoints );
CvMat c_objPt = objPt, c_imgPt = imgPt, c_npoints = npoints;
CvMat c_cameraMatrix = cameraMatrix, c_distCoeffs = distCoeffs;
CvMat c_rvecM = rvecM, c_tvecM = tvecM;
CvMat c_rvecM = rvecM, c_tvecM = tvecM, c_stdDev = stdDeviationsM, c_errors = errorsM;
double reprojErr = cvCalibrateCamera2(&c_objPt, &c_imgPt, &c_npoints, imageSize,
double reprojErr = cvCalibrateCamera2Internal(&c_objPt, &c_imgPt, &c_npoints, imageSize,
&c_cameraMatrix, &c_distCoeffs,
rvecs_needed ? &c_rvecM : NULL,
tvecs_needed ? &c_tvecM : NULL, flags, criteria );
tvecs_needed ? &c_tvecM : NULL,
stddev_needed ? &c_stdDev : NULL,
errors_needed ? &c_errors : NULL, flags, criteria );
if( stddev_needed )
{
stdDeviationsIntrinsics.create(CV_CALIB_NINTRINSIC, 1, CV_64F);
Mat stdDeviationsIntrinsicsMat = stdDeviationsIntrinsics.getMat();
std::memcpy(stdDeviationsIntrinsicsMat.ptr(), stdDeviationsM.ptr(),
CV_CALIB_NINTRINSIC*sizeof(double));
}
if ( stddev_ext_needed )
{
stdDeviationsExtrinsics.create(nimages*6, 1, CV_64F);
Mat stdDeviationsExtrinsicsMat = stdDeviationsExtrinsics.getMat();
std::memcpy(stdDeviationsExtrinsicsMat.ptr(),
stdDeviationsM.ptr() + CV_CALIB_NINTRINSIC*sizeof(double),
nimages*6*sizeof(double));
}
// overly complicated and inefficient rvec/ tvec handling to support vector<Mat>
for(int i = 0; i < nimages; i++ )

2
modules/calib3d/src/compat_ptsetreg.cpp
View File

@@ -241,6 +241,8 @@ bool CvLevMarq::updateAlt( const CvMat*& _param, CvMat*& _JtJ, CvMat*& _JtErr, d
cvNorm(param, prevParam, CV_RELATIVE_L2) < criteria.epsilon )
{
_param = param;
_JtJ = JtJ;
_JtErr = JtErr;
state = DONE;
return false;
}

92
modules/calib3d/test/test_cameracalibration.cpp
View File

@@ -259,7 +259,7 @@ protected:
virtual void calibrate( int imageCount, int* pointCounts,
CvSize imageSize, CvPoint2D64f* imagePoints, CvPoint3D64f* objectPoints,
double* distortionCoeffs, double* cameraMatrix, double* translationVectors,
double* rotationMatrices, int flags ) = 0;
double* rotationMatrices, double *stdDevs, double* perViewErrors, int flags ) = 0;
virtual void project( int pointCount, CvPoint3D64f* objectPoints,
double* rotationMatrix, double* translationVector,
double* cameraMatrix, double* distortion, CvPoint2D64f* imagePoints ) = 0;
@@ -303,9 +303,13 @@ void CV_CameraCalibrationTest::run( int start_from )
double* transVects;
double* rotMatrs;
double* stdDevs;
double* perViewErrors;
double* goodTransVects;
double* goodRotMatrs;
double* goodPerViewErrors;
double* goodStdDevs;
double cameraMatrix[3*3];
double distortion[5]={0,0,0,0,0};
@@ -424,9 +428,13 @@ void CV_CameraCalibrationTest::run( int start_from )
/* Allocate memory for translate vectors and rotmatrixs*/
transVects = (double*)cvAlloc(3 * 1 * numImages * sizeof(double));
rotMatrs = (double*)cvAlloc(3 * 3 * numImages * sizeof(double));
stdDevs = (double*)cvAlloc((CV_CALIB_NINTRINSIC + 6*numImages) * sizeof(double));
perViewErrors = (double*)cvAlloc(numImages * sizeof(double));
goodTransVects = (double*)cvAlloc(3 * 1 * numImages * sizeof(double));
goodRotMatrs = (double*)cvAlloc(3 * 3 * numImages * sizeof(double));
goodPerViewErrors = (double*)cvAlloc(numImages * sizeof(double));
goodStdDevs = (double*)cvAlloc((CV_CALIB_NINTRINSIC + 6*numImages) * sizeof(double));
/* Read object points */
i = 0;/* shift for current point */
@@ -501,6 +509,13 @@ void CV_CameraCalibrationTest::run( int start_from )
}
}
/* Read good stdDeviations */
for (i = 0; i < CV_CALIB_NINTRINSIC + numImages*6; i++)
{
values_read = fscanf(file, "%lf", goodStdDevs + i);
CV_Assert(values_read == 1);
}
calibFlags = 0
// + CV_CALIB_FIX_PRINCIPAL_POINT
// + CV_CALIB_ZERO_TANGENT_DIST
@@ -526,6 +541,8 @@ void CV_CameraCalibrationTest::run( int start_from )
cameraMatrix,
transVects,
rotMatrs,
stdDevs,
perViewErrors,
calibFlags );
/* ---- Reproject points to the image ---- */
@@ -553,6 +570,8 @@ void CV_CameraCalibrationTest::run( int start_from )
meanDy = 0;
for( currImage = 0; currImage < numImages; currImage++ )
{
double imageMeanDx = 0;
double imageMeanDy = 0;
for( currPoint = 0; currPoint < etalonSize.width * etalonSize.height; currPoint++ )
{
rx = reprojectPoints[i].x;
@@ -563,6 +582,9 @@ void CV_CameraCalibrationTest::run( int start_from )
meanDx += dx;
meanDy += dy;
imageMeanDx += dx*dx;
imageMeanDy += dy*dy;
dx = fabs(dx);
dy = fabs(dy);
@@ -573,6 +595,13 @@ void CV_CameraCalibrationTest::run( int start_from )
maxDy = dy;
i++;
}
goodPerViewErrors[currImage] = sqrt( (imageMeanDx + imageMeanDy) /
(etalonSize.width * etalonSize.height));
//only for c-version of test (it does not provides evaluation of perViewErrors
//and returns zeros)
if(perViewErrors[currImage] == 0.0)
perViewErrors[currImage] = goodPerViewErrors[currImage];
}
meanDx /= numImages * etalonSize.width * etalonSize.height;
@@ -613,6 +642,23 @@ void CV_CameraCalibrationTest::run( int start_from )
if( code < 0 )
goto _exit_;
/* ----- Compare per view re-projection errors ----- */
code = compare(perViewErrors,goodPerViewErrors, numImages,0.1,"per view errors vector");
if( code < 0 )
goto _exit_;
/* ----- Compare standard deviations of parameters ----- */
//only for c-version of test (it does not provides evaluation of stdDevs
//and returns zeros)
for ( i = 0; i < CV_CALIB_NINTRINSIC + 6*numImages; i++)
{
if(stdDevs[i] == 0.0)
stdDevs[i] = goodStdDevs[i];
}
code = compare(stdDevs,goodStdDevs, CV_CALIB_NINTRINSIC + 6*numImages,.5,"stdDevs vector");
if( code < 0 )
goto _exit_;
if( maxDx > 1.0 )
{
ts->printf( cvtest::TS::LOG,
@@ -636,8 +682,12 @@ void CV_CameraCalibrationTest::run( int start_from )
cvFree(&transVects);
cvFree(&rotMatrs);
cvFree(&stdDevs);
cvFree(&perViewErrors);
cvFree(&goodTransVects);
cvFree(&goodRotMatrs);
cvFree(&goodPerViewErrors);
cvFree(&goodStdDevs);
fclose(file);
file = 0;
@@ -676,20 +726,28 @@ protected:
virtual void calibrate( int imageCount, int* pointCounts,
CvSize imageSize, CvPoint2D64f* imagePoints, CvPoint3D64f* objectPoints,
double* distortionCoeffs, double* cameraMatrix, double* translationVectors,
double* rotationMatrices, int flags );
double* rotationMatrices, double *stdDevs, double* perViewErrors, int flags );
virtual void project( int pointCount, CvPoint3D64f* objectPoints,
double* rotationMatrix, double* translationVector,
double* cameraMatrix, double* distortion, CvPoint2D64f* imagePoints );
};
void CV_CameraCalibrationTest_C::calibrate( int imageCount, int* pointCounts,
void CV_CameraCalibrationTest_C::calibrate(int imageCount, int* pointCounts,
CvSize imageSize, CvPoint2D64f* imagePoints, CvPoint3D64f* objectPoints,
double* distortionCoeffs, double* cameraMatrix, double* translationVectors,
double* rotationMatrices, int flags )
double* rotationMatrices, double *stdDevs, double *perViewErrors, int flags )
{
int i, total = 0;
for( i = 0; i < imageCount; i++ )
{
perViewErrors[i] = 0.0;
total += pointCounts[i];
}
for( i = 0; i < CV_CALIB_NINTRINSIC + imageCount*6; i++)
{
stdDevs[i] = 0.0;
}
CvMat _objectPoints = cvMat(1, total, CV_64FC3, objectPoints);
CvMat _imagePoints = cvMat(1, total, CV_64FC2, imagePoints);
@@ -700,8 +758,7 @@ void CV_CameraCalibrationTest_C::calibrate( int imageCount, int* pointCounts,
CvMat _translationVectors = cvMat(imageCount, 3, CV_64F, translationVectors);
cvCalibrateCamera2(&_objectPoints, &_imagePoints, &_pointCounts, imageSize,
&_cameraMatrix, &_distCoeffs, &_rotationMatrices, &_translationVectors,
flags);
&_cameraMatrix, &_distCoeffs, &_rotationMatrices, &_translationVectors, flags);
}
void CV_CameraCalibrationTest_C::project( int pointCount, CvPoint3D64f* objectPoints,
@@ -728,22 +785,24 @@ protected:
virtual void calibrate( int imageCount, int* pointCounts,
CvSize imageSize, CvPoint2D64f* imagePoints, CvPoint3D64f* objectPoints,
double* distortionCoeffs, double* cameraMatrix, double* translationVectors,
double* rotationMatrices, int flags );
double* rotationMatrices, double *stdDevs, double* perViewErrors, int flags );
virtual void project( int pointCount, CvPoint3D64f* objectPoints,
double* rotationMatrix, double* translationVector,
double* cameraMatrix, double* distortion, CvPoint2D64f* imagePoints );
};
void CV_CameraCalibrationTest_CPP::calibrate( int imageCount, int* pointCounts,
void CV_CameraCalibrationTest_CPP::calibrate(int imageCount, int* pointCounts,
CvSize _imageSize, CvPoint2D64f* _imagePoints, CvPoint3D64f* _objectPoints,
double* _distortionCoeffs, double* _cameraMatrix, double* translationVectors,
double* rotationMatrices, int flags )
double* rotationMatrices, double *stdDevs, double *perViewErrors, int flags )
{
vector<vector<Point3f> > objectPoints( imageCount );
vector<vector<Point2f> > imagePoints( imageCount );
Size imageSize = _imageSize;
Mat cameraMatrix, distCoeffs(1,4,CV_64F,Scalar::all(0));
vector<Mat> rvecs, tvecs;
Mat stdDevsMatInt, stdDevsMatExt;
Mat perViewErrorsMat;
CvPoint3D64f* op = _objectPoints;
CvPoint2D64f* ip = _imagePoints;
@@ -770,8 +829,23 @@ void CV_CameraCalibrationTest_CPP::calibrate( int imageCount, int* pointCounts,
distCoeffs,
rvecs,
tvecs,
stdDevsMatInt,
stdDevsMatExt,
perViewErrorsMat,
flags );
assert( stdDevsMatInt.type() == CV_64F );
assert( stdDevsMatInt.total() == static_cast<size_t>(CV_CALIB_NINTRINSIC) );
memcpy( stdDevs, stdDevsMatInt.ptr(), CV_CALIB_NINTRINSIC*sizeof(double) );
assert( stdDevsMatExt.type() == CV_64F );
assert( stdDevsMatExt.total() == static_cast<size_t>(6*imageCount) );
memcpy( stdDevs + CV_CALIB_NINTRINSIC, stdDevsMatExt.ptr(), 6*imageCount*sizeof(double) );
assert( perViewErrorsMat.type() == CV_64F);
assert( perViewErrorsMat.total() == static_cast<size_t>(imageCount) );
memcpy( perViewErrors, perViewErrorsMat.ptr(), imageCount*sizeof(double) );
assert( cameraMatrix.type() == CV_64FC1 );
memcpy( _cameraMatrix, cameraMatrix.ptr(), 9*sizeof(double) );

2
modules/core/CMakeLists.txt
View File

@@ -1,6 +1,6 @@
set(the_description "The Core Functionality")
ocv_add_module(core
PRIVATE_REQUIRED ${ZLIB_LIBRARIES} "${OPENCL_LIBRARIES}" "${VA_LIBRARIES}"
PRIVATE_REQUIRED ${ZLIB_LIBRARIES} "${OPENCL_LIBRARIES}" "${VA_LIBRARIES}" "${OPENCV_HAL_LINKER_LIBS}"
OPTIONAL opencv_cudev
WRAP java python)

26
modules/core/include/opencv2/core/hal/hal.hpp
View File

@@ -49,17 +49,6 @@
#include "opencv2/core/cvstd.hpp"
#include "opencv2/core/hal/interface.h"
//! @cond IGNORED
#define CALL_HAL(name, fun, ...) \
int res = fun(__VA_ARGS__); \
if (res == CV_HAL_ERROR_OK) \
return; \
else if (res != CV_HAL_ERROR_NOT_IMPLEMENTED) \
CV_Error_(cv::Error::StsInternal, \
("HAL implementation " CVAUX_STR(name) " ==> " CVAUX_STR(fun) " returned %d (0x%08x)", res, res));
//! @endcond
namespace cv { namespace hal {
//! @addtogroup core_hal_functions
@@ -75,6 +64,21 @@ CV_EXPORTS int LU32f(float* A, size_t astep, int m, float* b, size_t bstep, int
CV_EXPORTS int LU64f(double* A, size_t astep, int m, double* b, size_t bstep, int n);
CV_EXPORTS bool Cholesky32f(float* A, size_t astep, int m, float* b, size_t bstep, int n);
CV_EXPORTS bool Cholesky64f(double* A, size_t astep, int m, double* b, size_t bstep, int n);
CV_EXPORTS void SVD32f(float* At, size_t astep, float* W, float* U, size_t ustep, float* Vt, size_t vstep, int m, int n, int flags);
CV_EXPORTS void SVD64f(double* At, size_t astep, double* W, double* U, size_t ustep, double* Vt, size_t vstep, int m, int n, int flags);
CV_EXPORTS void gemm32f(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags);
CV_EXPORTS void gemm64f(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags);
CV_EXPORTS void gemm32fc(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags);
CV_EXPORTS void gemm64fc(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags);
CV_EXPORTS int normL1_(const uchar* a, const uchar* b, int n);
CV_EXPORTS float normL1_(const float* a, const float* b, int n);

15
modules/core/include/opencv2/core/hal/interface.h
View File

@@ -158,6 +158,21 @@ typedef signed char schar;
#define CV_HAL_DFT_IS_INPLACE 1024
//! @}
//! @name SVD flags
//! @{
#define CV_HAL_SVD_NO_UV 1
#define CV_HAL_SVD_SHORT_UV 2
#define CV_HAL_SVD_MODIFY_A 4
#define CV_HAL_SVD_FULL_UV 8
//! @}
//! @name Gemm flags
//! @{
#define CV_HAL_GEMM_1_T 1
#define CV_HAL_GEMM_2_T 2
#define CV_HAL_GEMM_3_T 4
//! @}
//! @}
#endif

4
modules/core/include/opencv2/core/mat.hpp
View File

@@ -2880,9 +2880,9 @@ public:
//! copy operator
MatConstIterator_& operator = (const MatConstIterator_& it);
//! returns the current matrix element
_Tp operator *() const;
const _Tp& operator *() const;
//! returns the i-th matrix element, relative to the current
_Tp operator [](ptrdiff_t i) const;
const _Tp& operator [](ptrdiff_t i) const;
//! shifts the iterator forward by the specified number of elements
MatConstIterator_& operator += (ptrdiff_t ofs);

4
modules/core/include/opencv2/core/mat.inl.hpp
View File

@@ -2550,7 +2550,7 @@ MatConstIterator_<_Tp>& MatConstIterator_<_Tp>::operator = (const MatConstIterat
}
template<typename _Tp> inline
_Tp MatConstIterator_<_Tp>::operator *() const
const _Tp& MatConstIterator_<_Tp>::operator *() const
{
return *(_Tp*)(this->ptr);
}
@@ -2656,7 +2656,7 @@ MatConstIterator_<_Tp> operator - (const MatConstIterator_<_Tp>& a, ptrdiff_t of
}
template<typename _Tp> inline
_Tp MatConstIterator_<_Tp>::operator [](ptrdiff_t i) const
const _Tp& MatConstIterator_<_Tp>::operator [](ptrdiff_t i) const
{
return *(_Tp*)MatConstIterator::operator [](i);
}

2
modules/core/include/opencv2/core/matx.hpp
View File

@@ -438,7 +438,7 @@ template<typename _Tp, int m> struct Matx_DetOp
return p;
for( int i = 0; i < m; i++ )
p *= temp(i, i);
return 1./p;
return p;
}
};

485
modules/core/src/hal_internal.cpp Normal file
View File

@@ -0,0 +1,485 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Copyright (C) 2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "hal_internal.hpp"
#ifdef HAVE_LAPACK
#include <cmath>
#include <lapacke.h>
#include <cblas.h>
#include <algorithm>
#include <typeinfo>
#include <limits>
#include <complex>
#define HAL_GEMM_SMALL_COMPLEX_MATRIX_THRESH 100
#define HAL_GEMM_SMALL_MATRIX_THRESH 100
#define HAL_SVD_SMALL_MATRIX_THRESH 25
#define HAL_LU_SMALL_MATRIX_THRESH 100
#define HAL_CHOLESKY_SMALL_MATRIX_THRESH 100
//lapack stores matrices in column-major order so transposing is neded everywhere
template <typename fptype> static inline void
transpose_square_inplace(fptype *src, size_t src_ld, size_t m)
{
for(size_t i = 0; i < m - 1; i++)
for(size_t j = i + 1; j < m; j++)
std::swap(src[j*src_ld + i], src[i*src_ld + j]);
}
template <typename fptype> static inline void
transpose(const fptype *src, size_t src_ld, fptype* dst, size_t dst_ld, size_t m, size_t n)
{
for(size_t i = 0; i < m; i++)
for(size_t j = 0; j < n; j++)
dst[j*dst_ld + i] = src[i*src_ld + j];
}
template <typename fptype> static inline void
copy_matrix(const fptype *src, size_t src_ld, fptype* dst, size_t dst_ld, size_t m, size_t n)
{
for(size_t i = 0; i < m; i++)
for(size_t j = 0; j < n; j++)
dst[i*dst_ld + j] = src[i*src_ld + j];
}
template <typename fptype> static inline void
set_value(fptype *dst, size_t dst_ld, fptype value, size_t m, size_t n)
{
for(size_t i = 0; i < m; i++)
for(size_t j = 0; j < n; j++)
dst[i*dst_ld + j] = value;
}
template <typename fptype> static inline int
lapack_LU(fptype* a, size_t a_step, int m, fptype* b, size_t b_step, int n, int* info)
{
int lda = a_step / sizeof(fptype), sign = 0;
int* piv = new int[m];
transpose_square_inplace(a, lda, m);
if(b)
{
if(n == 1 && b_step == sizeof(fptype))
{
if(typeid(fptype) == typeid(float))
sgesv_(&m, &n, (float*)a, &lda, piv, (float*)b, &m, info);
else if(typeid(fptype) == typeid(double))
dgesv_(&m, &n, (double*)a, &lda, piv, (double*)b, &m, info);
}
else
{
int ldb = b_step / sizeof(fptype);
fptype* tmpB = new fptype[m*n];
transpose(b, ldb, tmpB, m, m, n);
if(typeid(fptype) == typeid(float))
sgesv_(&m, &n, (float*)a, &lda, piv, (float*)tmpB, &m, info);
else if(typeid(fptype) == typeid(double))
dgesv_(&m, &n, (double*)a, &lda, piv, (double*)tmpB, &m, info);
transpose(tmpB, m, b, ldb, n, m);
delete[] tmpB;
}
}
else
{
if(typeid(fptype) == typeid(float))
sgetrf_(&m, &m, (float*)a, &lda, piv, info);
else if(typeid(fptype) == typeid(double))
dgetrf_(&m, &m, (double*)a, &lda, piv, info);
}
if(*info == 0)
{
for(int i = 0; i < m; i++)
sign ^= piv[i] != i + 1;
*info = sign ? -1 : 1;
}
else
*info = 0; //in opencv LU function zero means error
delete[] piv;
return CV_HAL_ERROR_OK;
}
template <typename fptype> static inline int
lapack_Cholesky(fptype* a, size_t a_step, int m, fptype* b, size_t b_step, int n, bool* info)
{
int lapackStatus;
int lda = a_step / sizeof(fptype);
char L[] = {'L', '\0'};
if(b)
{
if(n == 1 && b_step == sizeof(fptype))
{
if(typeid(fptype) == typeid(float))
sposv_(L, &m, &n, (float*)a, &lda, (float*)b, &m, &lapackStatus);
else if(typeid(fptype) == typeid(double))
dposv_(L, &m, &n, (double*)a, &lda, (double*)b, &m, &lapackStatus);
}
else
{
int ldb = b_step / sizeof(fptype);
fptype* tmpB = new fptype[m*n];
transpose(b, ldb, tmpB, m, m, n);
if(typeid(fptype) == typeid(float))
sposv_(L, &m, &n, (float*)a, &lda, (float*)tmpB, &m, &lapackStatus);
else if(typeid(fptype) == typeid(double))
dposv_(L, &m, &n, (double*)a, &lda, (double*)tmpB, &m, &lapackStatus);
transpose(tmpB, m, b, ldb, n, m);
delete[] tmpB;
}
}
else
{
if(typeid(fptype) == typeid(float))
spotrf_(L, &m, (float*)a, &lda, &lapackStatus);
else if(typeid(fptype) == typeid(double))
dpotrf_(L, &m, (double*)a, &lda, &lapackStatus);
}
if(lapackStatus == 0) *info = true;
else *info = false; //in opencv Cholesky function false means error
return CV_HAL_ERROR_OK;
}
template <typename fptype> static inline int
lapack_SVD(fptype* a, size_t a_step, fptype *w, fptype* u, size_t u_step, fptype* vt, size_t v_step, int m, int n, int flags, int* info)
{
int lda = a_step / sizeof(fptype);
int ldv = v_step / sizeof(fptype);
int ldu = u_step / sizeof(fptype);
int lwork = -1;
int* iworkBuf = new int[8*std::min(m, n)];
fptype work1 = 0;
//A already transposed and m>=n
char mode[] = { ' ', '\0'};
if(flags & CV_HAL_SVD_NO_UV)
{
ldv = 1;
mode[0] = 'N';
}
else if((flags & CV_HAL_SVD_SHORT_UV) && (flags & CV_HAL_SVD_MODIFY_A)) //short SVD, U stored in a
mode[0] = 'O';
else if((flags & CV_HAL_SVD_SHORT_UV) && !(flags & CV_HAL_SVD_MODIFY_A)) //short SVD, U stored in u if m>=n
mode[0] = 'S';
else if(flags & CV_HAL_SVD_FULL_UV) //full SVD, U stored in u or in a
mode[0] = 'A';
if((flags & CV_HAL_SVD_MODIFY_A) && (flags & CV_HAL_SVD_FULL_UV)) //U stored in a
{
u = new fptype[m*m];
ldu = m;
}
if(typeid(fptype) == typeid(float))
sgesdd_(mode, &m, &n, (float*)a, &lda, (float*)w, (float*)u, &ldu, (float*)vt, &ldv, (float*)&work1, &lwork, iworkBuf, info);
else if(typeid(fptype) == typeid(double))
dgesdd_(mode, &m, &n, (double*)a, &lda, (double*)w, (double*)u, &ldu, (double*)vt, &ldv, (double*)&work1, &lwork, iworkBuf, info);
lwork = round(work1); //optimal buffer size
fptype* buffer = new fptype[lwork + 1];
if(typeid(fptype) == typeid(float))
sgesdd_(mode, &m, &n, (float*)a, &lda, (float*)w, (float*)u, &ldu, (float*)vt, &ldv, (float*)buffer, &lwork, iworkBuf, info);
else if(typeid(fptype) == typeid(double))
dgesdd_(mode, &m, &n, (double*)a, &lda, (double*)w, (double*)u, &ldu, (double*)vt, &ldv, (double*)buffer, &lwork, iworkBuf, info);
if(!(flags & CV_HAL_SVD_NO_UV))
transpose_square_inplace(vt, ldv, n);
if((flags & CV_HAL_SVD_MODIFY_A) && (flags & CV_HAL_SVD_FULL_UV))
{
for(int i = 0; i < m; i++)
for(int j = 0; j < m; j++)
a[i*lda + j] = u[i*m + j];
delete[] u;
}
delete[] iworkBuf;
delete[] buffer;
return CV_HAL_ERROR_OK;
}
template <typename fptype> static inline int
lapack_gemm(const fptype *src1, size_t src1_step, const fptype *src2, size_t src2_step, fptype alpha,
const fptype *src3, size_t src3_step, fptype beta, fptype *dst, size_t dst_step, int a_m, int a_n, int d_n, int flags)
{
int ldsrc1 = src1_step / sizeof(fptype);
int ldsrc2 = src2_step / sizeof(fptype);
int ldsrc3 = src3_step / sizeof(fptype);
int lddst = dst_step / sizeof(fptype);
int c_m, c_n, d_m;
CBLAS_TRANSPOSE transA, transB;
if(flags & CV_HAL_GEMM_2_T)
{
transB = CblasTrans;
if(flags & CV_HAL_GEMM_1_T )
{
d_m = a_n;
}
else
{
d_m = a_m;
}
}
else
{
transB = CblasNoTrans;
if(flags & CV_HAL_GEMM_1_T )
{
d_m = a_n;
}
else
{
d_m = a_m;
}
}
if(flags & CV_HAL_GEMM_3_T)
{
c_m = d_n;
c_n = d_m;
}
else
{
c_m = d_m;
c_n = d_n;
}
if(flags & CV_HAL_GEMM_1_T )
{
transA = CblasTrans;
std::swap(a_n, a_m);
}
else
{
transA = CblasNoTrans;
}
if(src3 != dst && beta != 0.0 && src3_step != 0) {
if(flags & CV_HAL_GEMM_3_T)
transpose(src3, ldsrc3, dst, lddst, c_m, c_n);
else
copy_matrix(src3, ldsrc3, dst, lddst, c_m, c_n);
}
else if (src3 == dst && (flags & CV_HAL_GEMM_3_T)) //actually transposing C in this case done by openCV
return CV_HAL_ERROR_NOT_IMPLEMENTED;
else if(src3_step == 0 && beta != 0.0)
set_value(dst, lddst, (fptype)0.0, d_m, d_n);
if(typeid(fptype) == typeid(float))
cblas_sgemm(CblasRowMajor, transA, transB, a_m, d_n, a_n, (float)alpha, (float*)src1, ldsrc1, (float*)src2, ldsrc2, (float)beta, (float*)dst, lddst);
else if(typeid(fptype) == typeid(double))
cblas_dgemm(CblasRowMajor, transA, transB, a_m, d_n, a_n, (double)alpha, (double*)src1, ldsrc1, (double*)src2, ldsrc2, (double)beta, (double*)dst, lddst);
return CV_HAL_ERROR_OK;
}
template <typename fptype> static inline int
lapack_gemm_c(const fptype *src1, size_t src1_step, const fptype *src2, size_t src2_step, fptype alpha,
const fptype *src3, size_t src3_step, fptype beta, fptype *dst, size_t dst_step, int a_m, int a_n, int d_n, int flags)
{
int ldsrc1 = src1_step / sizeof(std::complex<fptype>);
int ldsrc2 = src2_step / sizeof(std::complex<fptype>);
int ldsrc3 = src3_step / sizeof(std::complex<fptype>);
int lddst = dst_step / sizeof(std::complex<fptype>);
int c_m, c_n, d_m;
CBLAS_TRANSPOSE transA, transB;
std::complex<fptype> cAlpha(alpha, 0.0);
std::complex<fptype> cBeta(beta, 0.0);
if(flags & CV_HAL_GEMM_2_T)
{
transB = CblasTrans;
if(flags & CV_HAL_GEMM_1_T )
{
d_m = a_n;
}
else
{
d_m = a_m;
}
}
else
{
transB = CblasNoTrans;
if(flags & CV_HAL_GEMM_1_T )
{
d_m = a_n;
}
else
{
d_m = a_m;
}
}
if(flags & CV_HAL_GEMM_3_T)
{
c_m = d_n;
c_n = d_m;
}
else
{
c_m = d_m;
c_n = d_n;
}
if(flags & CV_HAL_GEMM_1_T )
{
transA = CblasTrans;
std::swap(a_n, a_m);
}
else
{
transA = CblasNoTrans;
}
if(src3 != dst && beta != 0.0 && src3_step != 0) {
if(flags & CV_HAL_GEMM_3_T)
transpose((std::complex<fptype>*)src3, ldsrc3, (std::complex<fptype>*)dst, lddst, c_m, c_n);
else
copy_matrix((std::complex<fptype>*)src3, ldsrc3, (std::complex<fptype>*)dst, lddst, c_m, c_n);
}
else if (src3 == dst && (flags & CV_HAL_GEMM_3_T)) //actually transposing C in this case done by openCV
return CV_HAL_ERROR_NOT_IMPLEMENTED;
else if(src3_step == 0 && beta != 0.0)
set_value((std::complex<fptype>*)dst, lddst, std::complex<fptype>(0.0, 0.0), d_m, d_n);
if(typeid(fptype) == typeid(float))
cblas_cgemm(CblasRowMajor, transA, transB, a_m, d_n, a_n, &cAlpha, (void*)src1, ldsrc1, (void*)src2, ldsrc2, &cBeta, (void*)dst, lddst);
else if(typeid(fptype) == typeid(double))
cblas_zgemm(CblasRowMajor, transA, transB, a_m, d_n, a_n, &cAlpha, (void*)src1, ldsrc1, (void*)src2, ldsrc2, &cBeta, (void*)dst, lddst);
return CV_HAL_ERROR_OK;
}
int lapack_LU32f(float* a, size_t a_step, int m, float* b, size_t b_step, int n, int* info)
{
if(m < HAL_LU_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_LU(a, a_step, m, b, b_step, n, info);
}
int lapack_LU64f(double* a, size_t a_step, int m, double* b, size_t b_step, int n, int* info)
{
if(m < HAL_LU_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_LU(a, a_step, m, b, b_step, n, info);
}
int lapack_Cholesky32f(float* a, size_t a_step, int m, float* b, size_t b_step, int n, bool *info)
{
if(m < HAL_CHOLESKY_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_Cholesky(a, a_step, m, b, b_step, n, info);
}
int lapack_Cholesky64f(double* a, size_t a_step, int m, double* b, size_t b_step, int n, bool *info)
{
if(m < HAL_CHOLESKY_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_Cholesky(a, a_step, m, b, b_step, n, info);
}
int lapack_SVD32f(float* a, size_t a_step, float *w, float* u, size_t u_step, float* vt, size_t v_step, int m, int n, int flags)
{
if(m < HAL_SVD_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
int info;
return lapack_SVD(a, a_step, w, u, u_step, vt, v_step, m, n, flags, &info);
}
int lapack_SVD64f(double* a, size_t a_step, double *w, double* u, size_t u_step, double* vt, size_t v_step, int m, int n, int flags)
{
if(m < HAL_SVD_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
int info;
return lapack_SVD(a, a_step, w, u, u_step, vt, v_step, m, n, flags, &info);
}
int lapack_gemm32f(const float *src1, size_t src1_step, const float *src2, size_t src2_step, float alpha,
const float *src3, size_t src3_step, float beta, float *dst, size_t dst_step, int m, int n, int k, int flags)
{
if(m < HAL_GEMM_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_gemm(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m, n, k, flags);
}
int lapack_gemm64f(const double *src1, size_t src1_step, const double *src2, size_t src2_step, double alpha,
const double *src3, size_t src3_step, double beta, double *dst, size_t dst_step, int m, int n, int k, int flags)
{
if(m < HAL_GEMM_SMALL_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_gemm(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m, n, k, flags);
}
int lapack_gemm32fc(const float *src1, size_t src1_step, const float *src2, size_t src2_step, float alpha,
const float *src3, size_t src3_step, float beta, float *dst, size_t dst_step, int m, int n, int k, int flags)
{
if(m < HAL_GEMM_SMALL_COMPLEX_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_gemm_c(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m, n, k, flags);
}
int lapack_gemm64fc(const double *src1, size_t src1_step, const double *src2, size_t src2_step, double alpha,
const double *src3, size_t src3_step, double beta, double *dst, size_t dst_step, int m, int n, int k, int flags)
{
if(m < HAL_GEMM_SMALL_COMPLEX_MATRIX_THRESH)
return CV_HAL_ERROR_NOT_IMPLEMENTED;
return lapack_gemm_c(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m, n, k, flags);
}
#endif //HAVE_LAPACK

96
modules/core/src/hal_internal.hpp Normal file
View File

@@ -0,0 +1,96 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Copyright (C) 2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef OPENCV_CORE_HAL_INTERNAL_HPP
#define OPENCV_CORE_HAL_INTERNAL_HPP
#include "precomp.hpp"
#ifdef HAVE_LAPACK
int lapack_LU32f(float* a, size_t a_step, int m, float* b, size_t b_step, int n, int* info);
int lapack_LU64f(double* a, size_t a_step, int m, double* b, size_t b_step, int n, int* info);
int lapack_Cholesky32f(float* a, size_t a_step, int m, float* b, size_t b_step, int n, bool* info);
int lapack_Cholesky64f(double* a, size_t a_step, int m, double* b, size_t b_step, int n, bool* info);
int lapack_SVD32f(float* a, size_t a_step, float* w, float* u, size_t u_step, float* vt, size_t v_step, int m, int n, int flags);
int lapack_SVD64f(double* a, size_t a_step, double* w, double* u, size_t u_step, double* vt, size_t v_step, int m, int n, int flags);
int lapack_gemm32f(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m, int n, int k, int flags);
int lapack_gemm64f(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m, int n, int k, int flags);
int lapack_gemm32fc(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m, int n, int k, int flags);
int lapack_gemm64fc(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m, int n, int k, int flags);
#undef cv_hal_LU32f
#define cv_hal_LU32f lapack_LU32f
#undef cv_hal_LU64f
#define cv_hal_LU64f lapack_LU64f
#undef cv_hal_Cholesky32f
#define cv_hal_Cholesky32f lapack_Cholesky32f
#undef cv_hal_Cholesky64f
#define cv_hal_Cholesky64f lapack_Cholesky64f
#undef cv_hal_SVD32f
#define cv_hal_SVD32f lapack_SVD32f
#undef cv_hal_SVD64f
#define cv_hal_SVD64f lapack_SVD64f
#undef cv_hal_gemm32f
#define cv_hal_gemm32f lapack_gemm32f
#undef cv_hal_gemm64f
#define cv_hal_gemm64f lapack_gemm64f
#undef cv_hal_gemm32fc
#define cv_hal_gemm32fc lapack_gemm32fc
#undef cv_hal_gemm64fc
#define cv_hal_gemm64fc lapack_gemm64fc
#endif //HAVE_LAPACK
#endif //OPENCV_CORE_HAL_INTERNAL_HPP

134
modules/core/src/hal_replacement.hpp
View File

@@ -472,14 +472,148 @@ inline int hal_ni_dctFree2D(cvhalDFT *context) { return CV_HAL_ERROR_NOT_IMPLEME
#define cv_hal_dctFree2D hal_ni_dctFree2D
//! @endcond
/**
Performs \f$LU\f$ decomposition of square matrix \f$A=P*L*U\f$ (where \f$P\f$ is permutation matrix) and solves matrix equation \f$A*X=B\f$.
Function returns the \f$sign\f$ of permutation \f$P\f$ via parameter info.
@param src1 pointer to input matrix \f$A\f$ stored in row major order. After finish of work src1 contains at least \f$U\f$ part of \f$LU\f$
decomposition which is appropriate for determainant calculation: \f$det(A)=sign*\prod_{j=1}^{M}a_{jj}\f$.
@param src1_step number of bytes each matrix \f$A\f$ row occupies.
@param m size of square matrix \f$A\f$.
@param src2 pointer to \f$M\times N\f$ matrix \f$B\f$ which is the right-hand side of system \f$A*X=B\f$. \f$B\f$ stored in row major order.
If src2 is null pointer only \f$LU\f$ decomposition will be performed. After finish of work src2 contains solution \f$X\f$ of system \f$A*X=B\f$.
@param src2_step number of bytes each matrix \f$B\f$ row occupies.
@param n number of right-hand vectors in \f$M\times N\f$ matrix \f$B\f$.
@param info indicates success of decomposition. If *info is equals to zero decomposition failed, othervise *info is equals to \f$sign\f$.
*/
//! @addtogroup core_hal_interface_decomp_lu LU matrix decomposition
//! @{
inline int hal_ni_LU32f(float* src1, size_t src1_step, int m, float* src2, size_t src2_step, int n, int* info) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_LU64f(double* src1, size_t src1_step, int m, double* src2, size_t src2_step, int n, int* info) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
//! @}
/**
Performs Cholesky decomposition of matrix \f$A = L*L^T\f$ and solves matrix equation \f$A*X=B\f$.
@param src1 pointer to input matrix \f$A\f$ stored in row major order. After finish of work src1 contains lower triangular matrix \f$L\f$.
@param src1_step number of bytes each matrix \f$A\f$ row occupies.
@param m size of square matrix \f$A\f$.
@param src2 pointer to \f$M\times N\f$ matrix \f$B\f$ which is the right-hand side of system \f$A*X=B\f$. B stored in row major order.
If src2 is null pointer only Cholesky decomposition will be performed. After finish of work src2 contains solution \f$X\f$ of system \f$A*X=B\f$.
@param src2_step number of bytes each matrix \f$B\f$ row occupies.
@param n number of right-hand vectors in \f$M\times N\f$ matrix \f$B\f$.
@param info indicates success of decomposition. If *info is false decomposition failed.
*/
//! @addtogroup core_hal_interface_decomp_cholesky Cholesky matrix decomposition
//! @{
inline int hal_ni_Cholesky32f(float* src1, size_t src1_step, int m, float* src2, size_t src2_step, int n, bool* info) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_Cholesky64f(double* src1, size_t src1_step, int m, double* src2, size_t src2_step, int n, bool* info) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
//! @}
/**
Performs singular value decomposition of \f$M\times N\f$(\f$M>N\f$) matrix \f$A = U*\Sigma*V^T\f$.
@param src pointer to input \f$M\times N\f$ matrix \f$A\f$ stored in column major order.
After finish of work src will be filled with rows of \f$U\f$ or not modified (depends of flag CV_HAL_SVD_MODIFY_A).
@param src_step number of bytes each matrix \f$A\f$ column occupies.
@param w pointer to array for singular values of matrix \f$A\f$ (i. e. first \f$N\f$ diagonal elements of matrix \f$\Sigma\f$).
@param u pointer to output \f$M\times N\f$ or \f$M\times M\f$ matrix \f$U\f$ (size depends of flags). Pointer must be valid if flag CV_HAL_SVD_MODIFY_A not used.
@param u_step number of bytes each matrix \f$U\f$ row occupies.
@param vt pointer to array for \f$N\times N\f$ matrix \f$V^T\f$.
@param vt_step number of bytes each matrix \f$V^T\f$ row occupies.
@param m number fo rows in matrix \f$A\f$.
@param n number of columns in matrix \f$A\f$.
@param flags algorithm options (combination of CV_HAL_SVD_FULL_UV, ...).
*/
//! @addtogroup core_hal_interface_decomp_svd Singular value matrix decomposition
//! @{
inline int hal_ni_SVD32f(float* src, size_t src_step, float* w, float* u, size_t u_step, float* vt, size_t vt_step, int m, int n, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_SVD64f(double* src, size_t src_step, double* w, double* u, size_t u_step, double* vt, size_t vt_step, int m, int n, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
//! @}
//! @cond IGNORED
#define cv_hal_LU32f hal_ni_LU32f
#define cv_hal_LU64f hal_ni_LU64f
#define cv_hal_Cholesky32f hal_ni_Cholesky32f
#define cv_hal_Cholesky64f hal_ni_Cholesky64f
#define cv_hal_SVD32f hal_ni_SVD32f
#define cv_hal_SVD64f hal_ni_SVD64f
//! @endcond
/**
The function performs generalized matrix multiplication similar to the gemm functions in BLAS level 3:
\f$D = \alpha*AB+\beta*C\f$
@param src1 pointer to input \f$M\times N\f$ matrix \f$A\f$ or \f$A^T\f$ stored in row major order.
@param src1_step number of bytes each matrix \f$A\f$ or \f$A^T\f$ row occupies.
@param src2 pointer to input \f$N\times K\f$ matrix \f$B\f$ or \f$B^T\f$ stored in row major order.
@param src2_step number of bytes each matrix \f$B\f$ or \f$B^T\f$ row occupies.
@param alpha \f$\alpha\f$ multiplier before \f$AB\f$
@param src3 pointer to input \f$M\times K\f$ matrix \f$C\f$ or \f$C^T\f$ stored in row major order.
@param src3_step number of bytes each matrix \f$C\f$ or \f$C^T\f$ row occupies.
@param beta \f$\beta\f$ multiplier before \f$C\f$
@param dst pointer to input \f$M\times K\f$ matrix \f$D\f$ stored in row major order.
@param dst_step number of bytes each matrix \f$D\f$ row occupies.
@param m number of rows in matrix \f$A\f$ or \f$A^T\f$, equals to number of rows in matrix \f$D\f$
@param n number of columns in matrix \f$A\f$ or \f$A^T\f$
@param k number of columns in matrix \f$B\f$ or \f$B^T\f$, equals to number of columns in matrix \f$D\f$
@param flags algorithm options (combination of CV_HAL_GEMM_1_T, ...).
*/
//! @addtogroup core_hal_interface_matrix_multiplication Matrix multiplication
//! @{
inline int hal_ni_gemm32f(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_gemm64f(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_gemm32fc(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_gemm64fc(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
//! @}
//! @cond IGNORED
#define cv_hal_gemm32f hal_ni_gemm32f
#define cv_hal_gemm64f hal_ni_gemm64f
#define cv_hal_gemm32fc hal_ni_gemm32fc
#define cv_hal_gemm64fc hal_ni_gemm64fc
//! @endcond
//! @}
#if defined __GNUC__
# pragma GCC diagnostic pop
#elif defined _MSC_VER
# pragma warning( pop )
#endif
#include "hal_internal.hpp"
#include "custom_hal.hpp"
//! @cond IGNORED
#define CALL_HAL_RET(name, fun, retval, ...) \
int res = fun(__VA_ARGS__, &retval); \
if (res == CV_HAL_ERROR_OK) \
return retval; \
else if (res != CV_HAL_ERROR_NOT_IMPLEMENTED) \
CV_Error_(cv::Error::StsInternal, \
("HAL implementation " CVAUX_STR(name) " ==> " CVAUX_STR(fun) " returned %d (0x%08x)", res, res));
#define CALL_HAL(name, fun, ...) \
int res = fun(__VA_ARGS__); \
if (res == CV_HAL_ERROR_OK) \
return; \
else if (res != CV_HAL_ERROR_NOT_IMPLEMENTED) \
CV_Error_(cv::Error::StsInternal, \
("HAL implementation " CVAUX_STR(name) " ==> " CVAUX_STR(fun) " returned %d (0x%08x)", res, res));
//! @endcond
#endif

37
modules/core/src/lapack.cpp
View File

@@ -570,11 +570,44 @@ JacobiSVDImpl_(_Tp* At, size_t astep, _Tp* _W, _Tp* Vt, size_t vstep,
static void JacobiSVD(float* At, size_t astep, float* W, float* Vt, size_t vstep, int m, int n, int n1=-1)
{
JacobiSVDImpl_(At, astep, W, Vt, vstep, m, n, !Vt ? 0 : n1 < 0 ? n : n1, FLT_MIN, FLT_EPSILON*2);
hal::SVD32f(At, astep, W, NULL, astep, Vt, vstep, m, n, n1);
}
static void JacobiSVD(double* At, size_t astep, double* W, double* Vt, size_t vstep, int m, int n, int n1=-1)
{
hal::SVD64f(At, astep, W, NULL, astep, Vt, vstep, m, n, n1);
}
template <typename fptype> static inline int
decodeSVDParameters(const fptype* U, const fptype* Vt, int m, int n, int n1)
{
int halSVDFlag = 0;
if(Vt == NULL)
halSVDFlag = CV_HAL_SVD_NO_UV;
else if(n1 <= 0 || n1 == n)
{
halSVDFlag = CV_HAL_SVD_SHORT_UV;
if(U == NULL)
halSVDFlag |= CV_HAL_SVD_MODIFY_A;
}
else if(n1 == m)
{
halSVDFlag = CV_HAL_SVD_FULL_UV;
if(U == NULL)
halSVDFlag |= CV_HAL_SVD_MODIFY_A;
}
return halSVDFlag;
}
void hal::SVD32f(float* At, size_t astep, float* W, float* U, size_t ustep, float* Vt, size_t vstep, int m, int n, int n1)
{
CALL_HAL(SVD32f, cv_hal_SVD32f, At, astep, W, U, ustep, Vt, vstep, m, n, decodeSVDParameters(U, Vt, m, n, n1))
JacobiSVDImpl_(At, astep, W, Vt, vstep, m, n, !Vt ? 0 : n1 < 0 ? n : n1, FLT_MIN, FLT_EPSILON*2);
}
void hal::SVD64f(double* At, size_t astep, double* W, double* U, size_t ustep, double* Vt, size_t vstep, int m, int n, int n1)
{
CALL_HAL(SVD64f, cv_hal_SVD64f, At, astep, W, U, ustep, Vt, vstep, m, n, decodeSVDParameters(U, Vt, m, n, n1))
JacobiSVDImpl_(At, astep, W, Vt, vstep, m, n, !Vt ? 0 : n1 < 0 ? n : n1, DBL_MIN, DBL_EPSILON*10);
}
@@ -745,7 +778,6 @@ double cv::determinant( InputArray _mat )
{
for( int i = 0; i < rows; i++ )
result *= a.at<float>(i,i);
result = 1./result;
}
}
}
@@ -769,7 +801,6 @@ double cv::determinant( InputArray _mat )
{
for( int i = 0; i < rows; i++ )
result *= a.at<double>(i,i);
result = 1./result;
}
}
}

250
modules/core/src/matmul.cpp
View File

@@ -864,73 +864,39 @@ static bool ocl_gemm( InputArray matA, InputArray matB, double alpha,
return k.run(2, globalsize, block_size!=1 ? localsize : NULL, false);
}
#endif
}
void cv::gemm( InputArray matA, InputArray matB, double alpha,
InputArray matC, double beta, OutputArray _matD, int flags )
static void gemmImpl( Mat A, Mat B, double alpha,
Mat C, double beta, Mat D, int flags )
{
#ifdef HAVE_CLAMDBLAS
CV_OCL_RUN(ocl::haveAmdBlas() && matA.dims() <= 2 && matB.dims() <= 2 && matC.dims() <= 2 && _matD.isUMat() &&
matA.cols() > 20 && matA.rows() > 20 && matB.cols() > 20, // since it works incorrect for small sizes
ocl_gemm_amdblas(matA, matB, alpha, matC, beta, _matD, flags))
#endif
#ifdef HAVE_OPENCL
CV_OCL_RUN(_matD.isUMat() && matA.dims() <= 2 && matB.dims() <= 2 && matC.dims() <= 2,
ocl_gemm(matA, matB, alpha, matC, beta, _matD, flags))
#endif
const int block_lin_size = 128;
const int block_size = block_lin_size * block_lin_size;
static double zero[] = {0,0,0,0};
static float zerof[] = {0,0,0,0};
Mat A = matA.getMat(), B = matB.getMat(), C = beta != 0 ? matC.getMat() : Mat();
Size a_size = A.size(), d_size;
int i, len = 0, type = A.type();
CV_Assert( type == B.type() && (type == CV_32FC1 || type == CV_64FC1 || type == CV_32FC2 || type == CV_64FC2) );
switch( flags & (GEMM_1_T|GEMM_2_T) )
{
case 0:
d_size = Size( B.cols, a_size.height );
len = B.rows;
CV_Assert( a_size.width == len );
break;
case 1:
d_size = Size( B.cols, a_size.width );
len = B.rows;
CV_Assert( a_size.height == len );
break;
case 2:
d_size = Size( B.rows, a_size.height );
len = B.cols;
CV_Assert( a_size.width == len );
break;
case 3:
d_size = Size( B.rows, a_size.width );
len = B.cols;
CV_Assert( a_size.height == len );
break;
}
if( !C.empty() )
{
CV_Assert( C.type() == type &&
(((flags&GEMM_3_T) == 0 && C.rows == d_size.height && C.cols == d_size.width) ||
((flags&GEMM_3_T) != 0 && C.rows == d_size.width && C.cols == d_size.height)));
}
_matD.create( d_size.height, d_size.width, type );
Mat D = _matD.getMat();
if( (flags & GEMM_3_T) != 0 && C.data == D.data )
{
transpose( C, C );
flags &= ~GEMM_3_T;
}
if( flags == 0 && 2 <= len && len <= 4 && (len == d_size.width || len == d_size.height) )
{
if( type == CV_32F )
@@ -1194,8 +1160,7 @@ void cv::gemm( InputArray matA, InputArray matB, double alpha,
GEMMSingleMulFunc singleMulFunc;
GEMMBlockMulFunc blockMulFunc;
GEMMStoreFunc storeFunc;
Mat *matD = &D, tmat;
size_t tmat_size = 0;
Mat *matD = &D;
const uchar* Cdata = C.data;
size_t Cstep = C.data ? (size_t)C.step : 0;
AutoBuffer<uchar> buf;
@@ -1226,13 +1191,6 @@ void cv::gemm( InputArray matA, InputArray matB, double alpha,
storeFunc = (GEMMStoreFunc)GEMMStore_64fc;
}
if( D.data == A.data || D.data == B.data )
{
tmat_size = (size_t)d_size.width*d_size.height*CV_ELEM_SIZE(type);
// Allocate tmat later, once the size of buf is known
matD = &tmat;
}
if( (d_size.width == 1 || len == 1) && !(flags & GEMM_2_T) && B.isContinuous() )
{
b_step = d_size.width == 1 ? 0 : CV_ELEM_SIZE(type);
@@ -1306,10 +1264,6 @@ void cv::gemm( InputArray matA, InputArray matB, double alpha,
(d_size.width <= block_lin_size &&
d_size.height <= block_lin_size && len <= block_lin_size) )
{
if( tmat_size > 0 ) {
buf.allocate(tmat_size);
tmat = Mat(d_size.height, d_size.width, type, (uchar*)buf );
}
singleMulFunc( A.ptr(), A.step, B.ptr(), b_step, Cdata, Cstep,
matD->ptr(), matD->step, a_size, d_size, alpha, beta, flags );
}
@@ -1369,14 +1323,12 @@ void cv::gemm( InputArray matA, InputArray matB, double alpha,
flags &= ~GEMM_1_T;
}
buf.allocate(d_buf_size + b_buf_size + a_buf_size + tmat_size);
buf.allocate(d_buf_size + b_buf_size + a_buf_size);
d_buf = (uchar*)buf;
b_buf = d_buf + d_buf_size;
if( is_a_t )
a_buf = b_buf + b_buf_size;
if( tmat_size > 0 )
tmat = Mat(d_size.height, d_size.width, type, b_buf + b_buf_size + a_buf_size );
for( i = 0; i < d_size.height; i += di )
{
@@ -1455,12 +1407,200 @@ void cv::gemm( InputArray matA, InputArray matB, double alpha,
}
}
}
if( matD != &D )
matD->copyTo(D);
}
}
template <typename fptype>inline static void
callGemmImpl(const fptype *src1, size_t src1_step, const fptype *src2, size_t src2_step, fptype alpha,
const fptype *src3, size_t src3_step, fptype beta, fptype *dst, size_t dst_step, int m_a, int n_a, int n_d, int flags, int type)
{
CV_StaticAssert(GEMM_1_T == CV_HAL_GEMM_1_T, "Incompatible GEMM_1_T flag in HAL");
CV_StaticAssert(GEMM_2_T == CV_HAL_GEMM_2_T, "Incompatible GEMM_2_T flag in HAL");
CV_StaticAssert(GEMM_3_T == CV_HAL_GEMM_3_T, "Incompatible GEMM_3_T flag in HAL");
int b_m, b_n, c_m, c_n, m_d;
if(flags & GEMM_2_T)
{
b_m = n_d;
if(flags & GEMM_1_T )
{
b_n = m_a;
m_d = n_a;
}
else
{
b_n = n_a;
m_d = m_a;
}
}
else
{
b_n = n_d;
if(flags & GEMM_1_T )
{
b_m = m_a;
m_d = n_a;
}
else
{
m_d = m_a;
b_m = n_a;
}
}
if(flags & GEMM_3_T)
{
c_m = n_d;
c_n = m_d;
}
else
{
c_m = m_d;
c_n = n_d;
}
Mat A, B, C;
if(src1 != NULL)
A = Mat(m_a, n_a, type, (void*)src1, src1_step);
if(src2 != NULL)
B = Mat(b_m, b_n, type, (void*)src2, src2_step);
if(src3 != NULL && beta != 0.0)
C = Mat(c_m, c_n, type, (void*)src3, src3_step);
Mat D(m_d, n_d, type, (void*)dst, dst_step);
gemmImpl(A, B, alpha, C, beta, D, flags);
}
}
void cv::hal::gemm32f(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags)
{
CALL_HAL(gemm32f, cv_hal_gemm32f, src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags)
callGemmImpl(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags, CV_32F);
}
void cv::hal::gemm64f(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags)
{
CALL_HAL(gemm64f, cv_hal_gemm64f, src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags)
callGemmImpl(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags, CV_64F);
}
CV_EXPORTS void cv::hal::gemm32fc(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags)
{
CALL_HAL(gemm32fc, cv_hal_gemm32fc, src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags)
callGemmImpl(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags, CV_32FC2);
}
CV_EXPORTS void cv::hal::gemm64fc(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
int m_a, int n_a, int n_d, int flags)
{
CALL_HAL(gemm64fc, cv_hal_gemm64fc, src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags)
callGemmImpl(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m_a, n_a, n_d, flags, CV_64FC2);
}
void cv::gemm( InputArray matA, InputArray matB, double alpha,
InputArray matC, double beta, OutputArray _matD, int flags )
{
#ifdef HAVE_CLAMDBLAS
CV_OCL_RUN(ocl::haveAmdBlas() && matA.dims() <= 2 && matB.dims() <= 2 && matC.dims() <= 2 && _matD.isUMat() &&
matA.cols() > 20 && matA.rows() > 20 && matB.cols() > 20, // since it works incorrect for small sizes
ocl_gemm_amdblas(matA, matB, alpha, matC, beta, _matD, flags))
#endif
#ifdef HAVE_OPENCL
CV_OCL_RUN(_matD.isUMat() && matA.dims() <= 2 && matB.dims() <= 2 && matC.dims() <= 2,
ocl_gemm(matA, matB, alpha, matC, beta, _matD, flags))
#endif
Mat A = matA.getMat(), B = matB.getMat(), C = beta != 0.0 ? matC.getMat() : Mat();
Size a_size = A.size(), d_size;
int len = 0, type = A.type();
CV_Assert( type == B.type() && (type == CV_32FC1 || type == CV_64FC1 || type == CV_32FC2 || type == CV_64FC2) );
switch( flags & (GEMM_1_T|GEMM_2_T) )
{
case 0:
d_size = Size( B.cols, a_size.height );
len = B.rows;
CV_Assert( a_size.width == len );
break;
case 1:
d_size = Size( B.cols, a_size.width );
len = B.rows;
CV_Assert( a_size.height == len );
break;
case 2:
d_size = Size( B.rows, a_size.height );
len = B.cols;
CV_Assert( a_size.width == len );
break;
case 3:
d_size = Size( B.rows, a_size.width );
len = B.cols;
CV_Assert( a_size.height == len );
break;
}
if( !C.empty() )
{
CV_Assert( C.type() == type &&
(((flags&GEMM_3_T) == 0 && C.rows == d_size.height && C.cols == d_size.width) ||
((flags&GEMM_3_T) != 0 && C.rows == d_size.width && C.cols == d_size.height)));
}
_matD.create( d_size.height, d_size.width, type );
Mat D = _matD.getMat();
if( (flags & GEMM_3_T) != 0 && C.data == D.data )
{
transpose( C, C );
flags &= ~GEMM_3_T;
}
Mat *DProxyPtr = &D, DProxy;
if( D.data == A.data || D.data == B.data )
{
DProxy = Mat(d_size.height, d_size.width, D.type());
DProxyPtr = &DProxy;
}
if( type == CV_32FC1 )
hal::gemm32f(A.ptr<float>(), A.step, B.ptr<float>(), B.step, static_cast<float>(alpha),
C.ptr<float>(), C.step, static_cast<float>(beta),
DProxyPtr->ptr<float>(), DProxyPtr->step,
a_size.height, a_size.width, DProxyPtr->cols, flags);
else if( type == CV_64FC1 )
hal::gemm64f(A.ptr<double>(), A.step, B.ptr<double>(), B.step, alpha,
C.ptr<double>(), C.step, beta,
DProxyPtr->ptr<double>(), DProxyPtr->step,
a_size.height, a_size.width, DProxyPtr->cols, flags);
else if( type == CV_32FC2 )
hal::gemm32fc(A.ptr<float>(), A.step, B.ptr<float>(), B.step, static_cast<float>(alpha),
C.ptr<float>(), C.step, static_cast<float>(beta),
DProxyPtr->ptr<float>(), DProxyPtr->step,
a_size.height, a_size.width, DProxyPtr->cols, flags);
else
{
CV_Assert( type == CV_64FC2 );
hal::gemm64fc(A.ptr<double>(), A.step, B.ptr<double>(), B.step, alpha,
C.ptr<double>(), C.step, beta,
D.ptr<double>(), D.step,
a_size.height, a_size.width, DProxyPtr->cols, flags);
}
if(DProxyPtr != &D)
DProxyPtr->copyTo(D);
}
/****************************************************************************************\
* Transform *
\****************************************************************************************/

19
modules/core/src/matrix_decomp.cpp
View File

@@ -89,8 +89,6 @@ LUImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n, _Tp eps)
for( k = 0; k < n; k++ )
b[j*bstep + k] += alpha*b[i*bstep + k];
}
A[i*astep + i] = -d;
}
if( b )
@@ -101,7 +99,7 @@ LUImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n, _Tp eps)
_Tp s = b[i*bstep + j];
for( k = i+1; k < m; k++ )
s -= A[i*astep + k]*b[k*bstep + j];
b[i*bstep + j] = s*A[i*astep + i];
b[i*bstep + j] = s/A[i*astep + i];
}
}
@@ -111,13 +109,19 @@ LUImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n, _Tp eps)
int LU32f(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n, FLT_EPSILON*10);
int output;
CALL_HAL_RET(LU32f, cv_hal_LU32f, output, A, astep, m, b, bstep, n)
output = LUImpl(A, astep, m, b, bstep, n, FLT_EPSILON*10);
return output;
}
int LU64f(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n, DBL_EPSILON*100);
int output;
CALL_HAL_RET(LU64f, cv_hal_LU64f, output, A, astep, m, b, bstep, n)
output = LUImpl(A, astep, m, b, bstep, n, DBL_EPSILON*100);
return output;
}
template<typename _Tp> static inline bool
@@ -193,14 +197,17 @@ CholImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n)
return true;
}
bool Cholesky32f(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
bool output;
CALL_HAL_RET(Cholesky32f, cv_hal_Cholesky32f, output, A, astep, m, b, bstep, n)
return CholImpl(A, astep, m, b, bstep, n);
}
bool Cholesky64f(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
bool output;
CALL_HAL_RET(Cholesky64f, cv_hal_Cholesky64f, output, A, astep, m, b, bstep, n)
return CholImpl(A, astep, m, b, bstep, n);
}

19
modules/imgproc/src/hal_replacement.hpp
View File

@@ -309,4 +309,23 @@ inline int hal_ni_warpPerspectve(int src_type, const uchar *src_data, size_t src
#include "custom_hal.hpp"
//! @cond IGNORED
#define CALL_HAL_RET(name, fun, retval, ...) \
int res = fun(__VA_ARGS__, &retval); \
if (res == CV_HAL_ERROR_OK) \
return retval; \
else if (res != CV_HAL_ERROR_NOT_IMPLEMENTED) \
CV_Error_(cv::Error::StsInternal, \
("HAL implementation " CVAUX_STR(name) " ==> " CVAUX_STR(fun) " returned %d (0x%08x)", res, res));
#define CALL_HAL(name, fun, ...) \
int res = fun(__VA_ARGS__); \
if (res == CV_HAL_ERROR_OK) \
return; \
else if (res != CV_HAL_ERROR_NOT_IMPLEMENTED) \
CV_Error_(cv::Error::StsInternal, \
("HAL implementation " CVAUX_STR(name) " ==> " CVAUX_STR(fun) " returned %d (0x%08x)", res, res));
//! @endcond
#endif

2
modules/video/include/opencv2/video/tracking.hpp
View File

@@ -226,7 +226,7 @@ CV_EXPORTS_W void calcOpticalFlowFarneback( InputArray prev, InputArray next, In
@param dst Second input 2D point set of the same size and the same type as A, or another image.
@param fullAffine If true, the function finds an optimal affine transformation with no additional
restrictions (6 degrees of freedom). Otherwise, the class of transformations to choose from is
limited to combinations of translation, rotation, and uniform scaling (5 degrees of freedom).
limited to combinations of translation, rotation, and uniform scaling (4 degrees of freedom).
The function finds an optimal affine transform *[A|b]* (a 2 x 3 floating-point matrix) that
approximates best the affine transformation between: