Point CylindricalWarperGpu::warp(const cuda::GpuMat & src, InputArray K, InputArray R, int interp_mode, int border_mode,
cuda::GpuMat & dst)
{
Rect dst_roi = buildMaps(src.size(), K, R, d_xmap_, d_ymap_);
dst.create(dst_roi.height + 1, dst_roi.width + 1, src.type());
cuda::remap(src, dst, d_xmap_, d_ymap_, interp_mode, border_mode);
return dst_roi.tl();
}
void build_psf_fft(int width, int height, float ffwhm_x)
{
PSFR = Mat::zeros(ImageR.size(), CV_32F);
int hs = fwhm_x * 0.3;
Mat kernelX = getGaussianKernel(hs * 10, ffwhm_x);
Mat kernelY = getGaussianKernel(hs * 10, ffwhm_x);
Mat kernel = kernelX * kernelY.t();
//Mat kernel;
//Mat kernel;
//resize(psf_file, kernel, Size(0, 0), 23.0/fwhm_x, 23.0/fwhm_x, INTER_AREA);
hs = kernel.cols;
int dy = height - hs;
int dx = width - hs;
pow(kernel, power/100.0, kernel);
imshow("kernel", kernel * 235);
copyMakeBorder(kernel, PSFR,
dy/2,
dy-(dy/2),
dx/2,
dx-(dx/2),
BORDER_CONSTANT, Scalar::all(0.0));
//imshow("kernel", PSFR);
PSFR.convertTo(PSFR, CV_32F);
PSFI = Mat::zeros(PSFR.size(), CV_32F);
Mat psf_planes[] = {PSFR, PSFI};
gPSFR.upload(PSFR);
gPSFI.upload(PSFI);
Scalar psum = cuda::sum(gPSFR);
cuda::multiply(gPSFR, 1.0 / psum[0], gPSFR);
psum = cuda::sum(gPSFR);
Mat complexPSF;
cuda::GpuMat gpPSF;
merge(psf_planes, 2, complexPSF);
gpPSF.upload(complexPSF);
cuda::dft(gpPSF, gpPSF, gpPSF.size());
gpPSF.download(complexPSF);
split(complexPSF, psf_planes);
//imshow("fft", PSFR);
}
TensorWrapper::TensorWrapper(cuda::GpuMat & mat, THCState *state) {
this->definedInLua = false;
if (mat.empty()) {
this->typeCode = CV_CUDA;
this->tensorPtr = nullptr;
return;
}
this->typeCode = CV_CUDA;
THCudaTensor *outputPtr = THCudaTensor_new(state);
// Build new storage on top of the Mat
outputPtr->storage = THCudaStorage_newWithData(
state,
reinterpret_cast<float *>(mat.data),
mat.step * mat.rows * mat.channels() / cv::getElemSize(mat.depth())
);
int sizeMultiplier;
if (mat.channels() == 1) {
outputPtr->nDimension = 2;
sizeMultiplier = cv::getElemSize(mat.depth());
} else {
outputPtr->nDimension = 3;
sizeMultiplier = mat.elemSize1();
}
outputPtr->size = static_cast<long *>(THAlloc(sizeof(long) * outputPtr->nDimension));
outputPtr->stride = static_cast<long *>(THAlloc(sizeof(long) * outputPtr->nDimension));
if (mat.channels() > 1) {
outputPtr->size[2] = mat.channels();
outputPtr->stride[2] = 1;
}
outputPtr->size[0] = mat.rows;
outputPtr->size[1] = mat.cols;
outputPtr->stride[0] = mat.step / sizeMultiplier;
outputPtr->stride[1] = mat.channels();
outputPtr->storageOffset = 0;
// Make OpenCV treat underlying data as user-allocated
mat.refcount = nullptr;
this->tensorPtr = reinterpret_cast<THByteTensor *>(outputPtr);
}
void CCubicGrids::extractRTFromBuffer(const cuda::GpuMat& cvgmSumBuf_, Eigen::Matrix3f* peimRw_, Eigen::Vector3f* peivTw_) const{
Mat cvmSumBuf;
cvgmSumBuf_.download(cvmSumBuf);
double* aHostTmp = (double*)cvmSumBuf.data;
//declare A and b
Eigen::Matrix<double, 6, 6, Eigen::RowMajor> A;
Eigen::Matrix<double, 6, 1> b;
//retrieve A and b from cvmSumBuf
short sShift = 0;
for (int i = 0; i < 6; ++i){ // rows
for (int j = i; j < 7; ++j) { // cols + b
double value = aHostTmp[sShift++];
if (j == 6) // vector b
b.data()[i] = value;
else
A.data()[j * 6 + i] = A.data()[i * 6 + j] = value;
}//for each col
}//for each row
//checking nullspace
double dDet = A.determinant();
if (fabs(dDet) < 1e-15 || dDet != dDet){
if (dDet != dDet)
PRINTSTR("Failure -- dDet cannot be qnan. ");
//reset ();
return;
}//if dDet is rational
//float maxc = A.maxCoeff();
Eigen::Matrix<float, 6, 1> result = A.llt().solve(b).cast<float>();
//Eigen::Matrix<float, 6, 1> result = A.jacobiSvd(ComputeThinU | ComputeThinV).solve(b);
float alpha = result(0);
float beta = result(1);
float gamma = result(2);
Eigen::Matrix3f Rinc = (Eigen::Matrix3f)Eigen::AngleAxisf(gamma, Eigen::Vector3f::UnitZ()) * Eigen::AngleAxisf(beta, Eigen::Vector3f::UnitY()) * Eigen::AngleAxisf(alpha, Eigen::Vector3f::UnitX());
Eigen::Vector3f tinc = result.tail<3>();
//compose
//eivTwCur = Rinc * eivTwCur + tinc;
//eimrmRwCur = Rinc * eimrmRwCur;
Eigen::Vector3f eivTinv = -peimRw_->transpose()* (*peivTw_);
Eigen::Matrix3f eimRinv = peimRw_->transpose();
eivTinv = Rinc * eivTinv + tinc;
eimRinv = Rinc * eimRinv;
*peivTw_ = -eimRinv.transpose() * eivTinv;
*peimRw_ = eimRinv.transpose();
}
inline
Mat::Mat(const cuda::GpuMat& m)
: flags(0), dims(0), rows(0), cols(0), data(0), datastart(0), dataend(0), datalimit(0), allocator(0), u(0), size(&rows)
{
m.download(*this);
}
void unmapFrame(cuda::GpuMat& frame)
{
cuSafeCall( cuvidUnmapVideoFrame(decoder_, (CUdeviceptr) frame.data) );
frame.release();
}
