Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

fix the bug when src*2 < dst

Browse Source
pull/22194/head
@lizhiyu3 2 years ago
parent 2920a8e0ec
commit 01226cb8ac
2 changed files with 167 additions and 10 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 140
      modules/imgproc/src/pyramids.cpp
  2. 37
      modules/imgproc/test/test_pyramid.cpp

140
modules/imgproc/src/pyramids.cpp
Unescape View File

@ -82,6 +82,8 @@ template<typename T1, typename T2> int PyrDownVecV(T1**, T2*, int) { return 0; }
template<typename T1, typename T2> int PyrUpVecV(T1**, T2**, int) { return 0; }
template<typename T1, typename T2> int PyrUpVecVOneRow(T1**, T2*, int) { return 0; }
#if CV_SIMD
template<> int PyrDownVecH<uchar, int, 1>(const uchar* src, int* row, int width)
@ -717,6 +719,120 @@ template <> int PyrUpVecV<float, float>(float** src, float** dst, int width)
return x;
}
template <> int PyrUpVecVOneRow<int, uchar>(int** src, uchar* dst, int width)
{
int x = 0;
const int *row0 = src[0], *row1 = src[1], *row2 = src[2];
for( ; x <= width - v_uint8::nlanes; x += v_uint8::nlanes)
{
v_int16 v_r00 = v_pack(vx_load(row0 + x), vx_load(row0 + x + v_int32::nlanes)),
v_r01 = v_pack(vx_load(row0 + x + 2 * v_int32::nlanes), vx_load(row0 + x + 3 * v_int32::nlanes)),
v_r10 = v_pack(vx_load(row1 + x), vx_load(row1 + x + v_int32::nlanes)),
v_r11 = v_pack(vx_load(row1 + x + 2 * v_int32::nlanes), vx_load(row1 + x + 3 * v_int32::nlanes)),
v_r20 = v_pack(vx_load(row2 + x), vx_load(row2 + x + v_int32::nlanes)),
v_r21 = v_pack(vx_load(row2 + x + 2 * v_int32::nlanes), vx_load(row2 + x + 3 * v_int32::nlanes));
v_int16 v_2r10 = v_r10 + v_r10, v_2r11 = (v_r11 + v_r11);
v_store(dst + x, v_rshr_pack_u<6>(v_r00 + v_r20 + (v_2r10 + v_2r10 + v_2r10), v_r01 + v_r21 + (v_2r11 + v_2r11 + v_2r11)));
}
if(x <= width - v_uint16::nlanes)
{
v_int16 v_r00 = v_pack(vx_load(row0 + x), vx_load(row0 + x + v_int32::nlanes)),
v_r10 = v_pack(vx_load(row1 + x), vx_load(row1 + x + v_int32::nlanes)),
v_r20 = v_pack(vx_load(row2 + x), vx_load(row2 + x + v_int32::nlanes));
v_int16 v_2r10 = v_r10 + v_r10;
v_rshr_pack_u_store<6>(dst + x, v_r00 + v_r20 + (v_2r10 + v_2r10 + v_2r10));
x += v_uint16::nlanes;
}
typedef int CV_DECL_ALIGNED(1) unaligned_int;
for (; x <= width - v_int32x4::nlanes; x += v_int32x4::nlanes)
{
v_int32 v_r00 = vx_load(row0 + x),
v_r10 = vx_load(row1 + x),
v_r20 = vx_load(row2 + x);
v_int32 v_2r10 = v_r10 + v_r10;
v_int16 d = v_pack(v_r00 + v_r20 + (v_2r10 + v_2r10 + v_2r10), (v_r10 + v_r20) << 2);
*(unaligned_int*)(dst + x) = v_reinterpret_as_s32(v_rshr_pack_u<6>(d, vx_setzero_s16())).get0();
}
vx_cleanup();
return x;
}
template <> int PyrUpVecVOneRow<int, short>(int** src, short* dst, int width)
{
int x = 0;
const int *row0 = src[0], *row1 = src[1], *row2 = src[2];
for( ; x <= width - v_int16::nlanes; x += v_int16::nlanes)
{
v_int32 v_r00 = vx_load(row0 + x),
v_r01 = vx_load(row0 + x + v_int32::nlanes),
v_r10 = vx_load(row1 + x),
v_r11 = vx_load(row1 + x + v_int32::nlanes),
v_r20 = vx_load(row2 + x),
v_r21 = vx_load(row2 + x + v_int32::nlanes);
v_store(dst + x, v_rshr_pack<6>(v_r00 + v_r20 + ((v_r10 << 1) + (v_r10 << 2)), v_r01 + v_r21 + ((v_r11 << 1) + (v_r11 << 2))));
}
if(x <= width - v_int32::nlanes)
{
v_int32 v_r00 = vx_load(row0 + x),
v_r10 = vx_load(row1 + x),
v_r20 = vx_load(row2 + x);
v_rshr_pack_store<6>(dst + x, v_r00 + v_r20 + ((v_r10 << 1) + (v_r10 << 2)));
x += v_int32::nlanes;
}
vx_cleanup();
return x;
}
template <> int PyrUpVecVOneRow<int, ushort>(int** src, ushort* dst, int width)
{
int x = 0;
const int *row0 = src[0], *row1 = src[1], *row2 = src[2];
for( ; x <= width - v_uint16::nlanes; x += v_uint16::nlanes)
{
v_int32 v_r00 = vx_load(row0 + x),
v_r01 = vx_load(row0 + x + v_int32::nlanes),
v_r10 = vx_load(row1 + x),
v_r11 = vx_load(row1 + x + v_int32::nlanes),
v_r20 = vx_load(row2 + x),
v_r21 = vx_load(row2 + x + v_int32::nlanes);
v_store(dst + x, v_rshr_pack_u<6>(v_r00 + v_r20 + ((v_r10 << 1) + (v_r10 << 2)), v_r01 + v_r21 + ((v_r11 << 1) + (v_r11 << 2))));
}
if(x <= width - v_int32::nlanes)
{
v_int32 v_r00 = vx_load(row0 + x),
v_r10 = vx_load(row1 + x),
v_r20 = vx_load(row2 + x);
v_rshr_pack_u_store<6>(dst + x, v_r00 + v_r20 + ((v_r10 << 1) + (v_r10 << 2)));
x += v_int32::nlanes;
}
vx_cleanup();
return x;
}
template <> int PyrUpVecVOneRow<float, float>(float** src, float* dst, int width)
{
int x = 0;
const float *row0 = src[0], *row1 = src[1], *row2 = src[2];
v_float32 v_6 = vx_setall_f32(6.0f), v_scale = vx_setall_f32(1.f/64.f);
for( ; x <= width - v_float32::nlanes; x += v_float32::nlanes)
{
v_float32 v_r0 = vx_load(row0 + x),
v_r1 = vx_load(row1 + x),
v_r2 = vx_load(row2 + x);
v_store(dst + x, v_scale * (v_muladd(v_6, v_r1, v_r0) + v_r2));
}
vx_cleanup();
return x;
}
#endif
template<class CastOp>
@ -963,7 +1079,7 @@ pyrUp_( const Mat& _src, Mat& _dst, int)
if (dsize.width > ssize.width*2)
{
row[(_dst.cols-1) * cn + x] = row[dx + cn];
row[(_dst.cols-1) + x] = row[dx + cn];
}
}
@ -983,12 +1099,24 @@ pyrUp_( const Mat& _src, Mat& _dst, int)
row0 = rows[0]; row1 = rows[1]; row2 = rows[2];
dsts[0] = dst0; dsts[1] = dst1;
x = PyrUpVecV<WT, T>(rows, dsts, dsize.width);
for( ; x < dsize.width; x++ )
if (dst0 != dst1)
{
T t1 = castOp((row1[x] + row2[x])*4);
T t0 = castOp(row0[x] + row1[x]*6 + row2[x]);
dst1[x] = t1; dst0[x] = t0;
x = PyrUpVecV<WT, T>(rows, dsts, dsize.width);
for( ; x < dsize.width; x++ )
{
T t1 = castOp((row1[x] + row2[x])*4);
T t0 = castOp(row0[x] + row1[x]*6 + row2[x]);
dst1[x] = t1; dst0[x] = t0;
}
}
else
{
x = PyrUpVecVOneRow<WT, T>(rows, dst0, dsize.width);
for( ; x < dsize.width; x++ )
{
T t0 = castOp(row0[x] + row1[x]*6 + row2[x]);
dst0[x] = t0;
}
}
}

37
modules/imgproc/test/test_pyramid.cpp
Unescape View File

@ -8,12 +8,41 @@ namespace opencv_test { namespace {
TEST(Imgproc_PyrUp, pyrUp_regression_22184)
{
Mat src(100, 100, CV_16UC3, Scalar::all(255));
Mat dst(100 * 2 + 1, 100 * 2 + 1, CV_16UC3, Scalar::all(0));
Mat src(100,100,CV_16UC3,Scalar(255,255,255));
Mat dst(100 * 2 + 1, 100 * 2 + 1, CV_16UC3, Scalar(0,0,0));
pyrUp(src, dst, Size(dst.cols, dst.rows));
double min_val = 0;
double min_val;
minMaxLoc(dst, &min_val);
ASSERT_GT(cvRound(min_val), 0);
}
}} // namespace
TEST(Imgproc_PyrUp, pyrUp_regression_22194)
{
Mat src(13, 13,CV_16UC3,Scalar(0,0,0));
{
int swidth = src.cols;
int sheight = src.rows;
int cn = src.channels();
int count = 0;
for (int y = 0; y < sheight; y++)
{
ushort *src_c = src.ptr<ushort>(y);
for (int x = 0; x < swidth * cn; x++)
{
src_c[x] = (count++) % 10;
}
}
}
Mat dst(src.cols * 2 - 1, src.rows * 2 - 1, CV_16UC3, Scalar(0,0,0));
pyrUp(src, dst, Size(dst.cols, dst.rows));
{
ushort *dst_c = dst.ptr<ushort>(dst.rows - 1);
ASSERT_EQ(dst_c[0], 6);
ASSERT_EQ(dst_c[1], 6);
ASSERT_EQ(dst_c[2], 1);
}
}
}
}

Write Preview
Loading…
Cancel
Save