fix cv::gpu::resize for INTER_LINEAR, now it produces the same result as CPU version

(cherry picked from commit da9be8231f)
pull/4080/head
Vladislav Vinogradov 10 years ago committed by Alexander Smorkalov
parent bb93c53948
commit 562796e41b
  1. 34
      modules/gpu/src/cuda/resize.cu
  2. 44
      modules/gpu/test/test_resize.cpp

@ -77,8 +77,8 @@ namespace cv { namespace gpu { namespace device
if (dst_x < dst.cols && dst_y < dst.rows)
{
const float src_x = dst_x * fx;
const float src_y = dst_y * fy;
const float src_x = (dst_x + 0.5f) * fx - 0.5f;
const float src_y = (dst_y + 0.5f) * fy - 0.5f;
work_type out = VecTraits<work_type>::all(0);
@ -86,16 +86,18 @@ namespace cv { namespace gpu { namespace device
const int y1 = __float2int_rd(src_y);
const int x2 = x1 + 1;
const int y2 = y1 + 1;
const int x2_read = ::min(x2, src.cols - 1);
const int y2_read = ::min(y2, src.rows - 1);
const int x1_read = ::max(::min(x1, src.cols - 1), 0);
const int y1_read = ::max(::min(y1, src.rows - 1), 0);
const int x2_read = ::max(::min(x2, src.cols - 1), 0);
const int y2_read = ::max(::min(y2, src.rows - 1), 0);
T src_reg = src(y1, x1);
T src_reg = src(y1_read, x1_read);
out = out + src_reg * ((x2 - src_x) * (y2 - src_y));
src_reg = src(y1, x2_read);
src_reg = src(y1_read, x2_read);
out = out + src_reg * ((src_x - x1) * (y2 - src_y));
src_reg = src(y2_read, x1);
src_reg = src(y2_read, x1_read);
out = out + src_reg * ((x2 - src_x) * (src_y - y1));
src_reg = src(y2_read, x2_read);
@ -119,6 +121,20 @@ namespace cv { namespace gpu { namespace device
}
}
template <class Ptr2D, typename T> __global__ void resize_linear(const Ptr2D src, PtrStepSz<T> dst, const float fy, const float fx)
{
const int dst_x = blockDim.x * blockIdx.x + threadIdx.x;
const int dst_y = blockDim.y * blockIdx.y + threadIdx.y;
if (dst_x < dst.cols && dst_y < dst.rows)
{
const float src_x = (dst_x + 0.5f) * fx - 0.5f;
const float src_y = (dst_y + 0.5f) * fy - 0.5f;
dst(dst_y, dst_x) = src(src_y, src_x);
}
}
template <typename Ptr2D, typename T> __global__ void resize_area(const Ptr2D src, PtrStepSz<T> dst)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
@ -231,7 +247,7 @@ namespace cv { namespace gpu { namespace device
TextureAccessor<T> texSrc = texAccessor(src, 0, 0);
LinearFilter< TextureAccessor<T> > filteredSrc(texSrc);
resize<<<grid, block>>>(filteredSrc, dst, fy, fx);
resize_linear<<<grid, block>>>(filteredSrc, dst, fy, fx);
}
else
{
@ -241,7 +257,7 @@ namespace cv { namespace gpu { namespace device
BorderReader<TextureAccessor<T>, BrdReplicate<T> > brdSrc(texSrc, brd);
LinearFilter< BorderReader<TextureAccessor<T>, BrdReplicate<T> > > filteredSrc(brdSrc);
resize<<<grid, block>>>(filteredSrc, dst, fy, fx);
resize_linear<<<grid, block>>>(filteredSrc, dst, fy, fx);
}
cudaSafeCall( cudaGetLastError() );

@ -73,6 +73,28 @@ namespace
}
}
template <typename T, template <typename> class Interpolator>
void resizeLinearImpl(const cv::Mat& src, cv::Mat& dst, double fx, double fy)
{
const int cn = src.channels();
cv::Size dsize(cv::saturate_cast<int>(src.cols * fx), cv::saturate_cast<int>(src.rows * fy));
dst.create(dsize, src.type());
float ifx = static_cast<float>(1.0 / fx);
float ify = static_cast<float>(1.0 / fy);
for (int y = 0; y < dsize.height; ++y)
{
for (int x = 0; x < dsize.width; ++x)
{
for (int c = 0; c < cn; ++c)
dst.at<T>(y, x * cn + c) = Interpolator<T>::getValue(src, (y + 0.5f) * ify - 0.5f, (x + 0.5f) * ifx - 0.5f, c, cv::BORDER_REPLICATE);
}
}
}
void resizeGold(const cv::Mat& src, cv::Mat& dst, double fx, double fy, int interpolation)
{
typedef void (*func_t)(const cv::Mat& src, cv::Mat& dst, double fx, double fy);
@ -90,12 +112,12 @@ namespace
static const func_t linear_funcs[] =
{
resizeImpl<unsigned char, LinearInterpolator>,
resizeImpl<signed char, LinearInterpolator>,
resizeImpl<unsigned short, LinearInterpolator>,
resizeImpl<short, LinearInterpolator>,
resizeImpl<int, LinearInterpolator>,
resizeImpl<float, LinearInterpolator>
resizeLinearImpl<unsigned char, LinearInterpolator>,
resizeLinearImpl<signed char, LinearInterpolator>,
resizeLinearImpl<unsigned short, LinearInterpolator>,
resizeLinearImpl<short, LinearInterpolator>,
resizeLinearImpl<int, LinearInterpolator>,
resizeLinearImpl<float, LinearInterpolator>
};
static const func_t cubic_funcs[] =
@ -204,7 +226,15 @@ INSTANTIATE_TEST_CASE_P(GPU_ImgProc, ResizeSameAsHost, testing::Combine(
DIFFERENT_SIZES,
testing::Values(MatType(CV_8UC1), MatType(CV_8UC3), MatType(CV_8UC4), MatType(CV_16UC1), MatType(CV_16UC3), MatType(CV_16UC4), MatType(CV_32FC1), MatType(CV_32FC3), MatType(CV_32FC4)),
testing::Values(0.3, 0.5),
testing::Values(Interpolation(cv::INTER_NEAREST), Interpolation(cv::INTER_AREA)),
testing::Values(Interpolation(cv::INTER_NEAREST), Interpolation(cv::INTER_LINEAR), Interpolation(cv::INTER_AREA)),
WHOLE_SUBMAT));
INSTANTIATE_TEST_CASE_P(GPU_ImgProc2, ResizeSameAsHost, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
testing::Values(MatType(CV_8UC1), MatType(CV_8UC3), MatType(CV_8UC4), MatType(CV_16UC1), MatType(CV_16UC3), MatType(CV_16UC4), MatType(CV_32FC1), MatType(CV_32FC3), MatType(CV_32FC4)),
testing::Values(0.3, 0.5, 1.5, 2.0),
testing::Values(Interpolation(cv::INTER_NEAREST), Interpolation(cv::INTER_LINEAR)),
WHOLE_SUBMAT));
#endif // HAVE_CUDA

Loading…
Cancel
Save