created wrappers for new NPP functions

removed void integral(const GpuMat& src, GpuMat& sum, GpuMat& sqsum, Stream& stream) - it fails with NPP_NOT_IMPLEMENTED error
updated docs, accuracy and performance tests
This commit is contained in:
Vladislav Vinogradov
2012-02-22 10:00:53 +00:00
parent e426dfc396
commit 2d30480982
37 changed files with 1984 additions and 566 deletions

View File

@@ -48,6 +48,7 @@ using namespace cv::gpu;
#if !defined (HAVE_CUDA)
void cv::gpu::meanStdDev(const GpuMat&, Scalar&, Scalar&) { throw_nogpu(); }
void cv::gpu::meanStdDev(const GpuMat&, Scalar&, Scalar&, GpuMat&) { throw_nogpu(); }
double cv::gpu::norm(const GpuMat&, int) { throw_nogpu(); return 0.0; }
double cv::gpu::norm(const GpuMat&, int, GpuMat&) { throw_nogpu(); return 0.0; }
double cv::gpu::norm(const GpuMat&, const GpuMat&, int) { throw_nogpu(); return 0.0; }
@@ -108,6 +109,12 @@ namespace
// meanStdDev
void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev)
{
GpuMat buf;
meanStdDev(src, mean, stddev, buf);
}
void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev, GpuMat& buf)
{
CV_Assert(src.type() == CV_8UC1);
@@ -117,15 +124,12 @@ void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev)
DeviceBuffer dbuf(2);
#if CUDART_VERSION > 4000
int bufSize;
nppSafeCall( nppiMeanStdDev8uC1RGetBufferHostSize(sz, &bufSize) );
GpuMat buf(1, bufSize, CV_8UC1);
ensureSizeIsEnough(1, bufSize, CV_8UC1, buf);
nppSafeCall( nppiMean_StdDev_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step), sz, buf.ptr<Npp8u>(), dbuf, (double*)dbuf + 1) );
#else
nppSafeCall( nppiMean_StdDev_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step), sz, dbuf, (double*)dbuf + 1) );
#endif
cudaSafeCall( cudaDeviceSynchronize() );
@@ -133,7 +137,6 @@ void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev)
dbuf.download(ptrs);
}
////////////////////////////////////////////////////////////////////////
// norm
@@ -151,7 +154,7 @@ double cv::gpu::norm(const GpuMat& src, int normType, GpuMat& buf)
return absSum(src_single_channel, buf)[0];
if (normType == NORM_L2)
return sqrt(sqrSum(src_single_channel, buf)[0]);
return std::sqrt(sqrSum(src_single_channel, buf)[0]);
if (normType == NORM_INF)
{