potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-26 22:03:04 +04:00
Code Packages Releases Wiki Activity
Files
next
opencv/modules/imgproc/test/test_resize_bitexact.cpp
T
Matt Van Horn 5e6a591357 Merge pull request #28873 from mvanhorn:osc/28429-fix-inter-nearest-exact-fp-5x 
imgproc: fix INTER_NEAREST_EXACT to match Pillow (5.x retarget of #28661) #28873

## Summary

Fixes `INTER_NEAREST_EXACT` producing different results from Pillow for images with dimensions that are multiples of 64 (e.g., 128, 192).

Retargets #28661 to 5.x per core team request.

## Root cause

The old 16-bit fixed-point arithmetic (`ifx`, `ifx0`, `>> 16`) rounds differently from Pillow's pixel-center coordinate mapping. At exact pixel boundaries (e.g., y=7 with 128 to 160: `0.4 + 7*0.8 = 5.999999999999999`), Pillow's truncation drops to 5 while the old opencv formula rounds to 6.

## Changes

- `resize.cpp`: Replace `ifx/ifx0/ify/ify0` 16-bit fixed-point with an int64 arithmetic formula equivalent to `floor((i + 0.5) * src / dst)`, computed as `((int64_t)(i * 2 + 1) * src) / (dst * 2)`. This matches Pillow's pixel-center mapping exactly, is deterministic across platforms, and avoids FP rounding divergence entirely.
- `test_resize_bitexact.cpp`: Add `NearestExact_PillowCompat` covering 4 dimension pairs including the reproducer from #28429.

## Testing

- All 3 `Resize_Bitexact` tests pass (Linear8U, Nearest8U, NearestExact_PillowCompat) on this branch built against 5.x
- Verified bit-exact match to PIL output (Python 3.14, Pillow) on 49 dimension combinations including random pairs
- Built on macOS ARM64 with HAL (carotene/KleidiCV) active

Fixes #28429
2026-06-05 19:35:44 +03:00

304 lines
13 KiB
C++
Raw Permalink Blame History

// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#include "test_precomp.hpp"
namespace opencv_test { namespace {
static const int fixedShiftU8 = 8;
template <typename T, int fixedShift>
void eval4(int64_t xcoeff0, int64_t xcoeff1, int64_t ycoeff0, int64_t ycoeff1, int cn,
uint8_t* src_pt00, uint8_t* src_pt01, uint8_t* src_pt10, uint8_t* src_pt11, uint8_t* dst_pt)
{
static const int64_t fixedRound = ((1LL << (fixedShift * 2)) >> 1);
int64_t val = (((T*)src_pt00)[cn] * xcoeff0 + ((T*)src_pt01)[cn] * xcoeff1) * ycoeff0 +
(((T*)src_pt10)[cn] * xcoeff0 + ((T*)src_pt11)[cn] * xcoeff1) * ycoeff1 ;
((T*)dst_pt)[cn] = saturate_cast<T>((val + fixedRound) >> (fixedShift * 2));
}
TEST(Resize_Bitexact, Linear8U)
{
static const int64_t fixedOne = (1L << fixedShiftU8);
struct testmode
{
int type;
Size sz;
} modes[] = {
{ CV_8UC1, Size( 512, 768) }, // 1/2 1
{ CV_8UC3, Size( 512, 768) },
{ CV_8UC1, Size(1024, 384) }, // 1 1/2
{ CV_8UC4, Size(1024, 384) },
{ CV_8UC1, Size( 512, 384) }, // 1/2 1/2
{ CV_8UC2, Size( 512, 384) },
{ CV_8UC3, Size( 512, 384) },
{ CV_8UC4, Size( 512, 384) },
{ CV_8UC1, Size( 256, 192) }, // 1/4 1/4
{ CV_8UC2, Size( 256, 192) },
{ CV_8UC3, Size( 256, 192) },
{ CV_8UC4, Size( 256, 192) },
{ CV_8UC1, Size( 4, 3) }, // 1/256 1/256
{ CV_8UC2, Size( 4, 3) },
{ CV_8UC3, Size( 4, 3) },
{ CV_8UC4, Size( 4, 3) },
{ CV_8UC1, Size( 342, 384) }, // 1/3 1/2
{ CV_8UC1, Size( 342, 256) }, // 1/3 1/3
{ CV_8UC2, Size( 342, 256) },
{ CV_8UC3, Size( 342, 256) },
{ CV_8UC4, Size( 342, 256) },
{ CV_8UC1, Size( 512, 256) }, // 1/2 1/3
{ CV_8UC1, Size( 146, 110) }, // 1/7 1/7
{ CV_8UC3, Size( 146, 110) },
{ CV_8UC4, Size( 146, 110) },
{ CV_8UC1, Size( 931, 698) }, // 10/11 10/11
{ CV_8UC2, Size( 931, 698) },
{ CV_8UC3, Size( 931, 698) },
{ CV_8UC4, Size( 931, 698) },
{ CV_8UC1, Size( 853, 640) }, // 10/12 10/12
{ CV_8UC3, Size( 853, 640) },
{ CV_8UC4, Size( 853, 640) },
{ CV_8UC1, Size(1004, 753) }, // 251/256 251/256
{ CV_8UC2, Size(1004, 753) },
{ CV_8UC3, Size(1004, 753) },
{ CV_8UC4, Size(1004, 753) },
{ CV_8UC1, Size(2048,1536) }, // 2 2
{ CV_8UC2, Size(2048,1536) },
{ CV_8UC4, Size(2048,1536) },
{ CV_8UC1, Size(3072,2304) }, // 3 3
{ CV_8UC3, Size(3072,2304) },
{ CV_8UC1, Size(7168,5376) } // 7 7
};
for (int modeind = 0, _modecnt = sizeof(modes) / sizeof(modes[0]); modeind < _modecnt; ++modeind)
{
int type = modes[modeind].type, depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
int dcols = modes[modeind].sz.width, drows = modes[modeind].sz.height;
int cols = 1024, rows = 768;
double inv_scale_x = (double)dcols / cols;
double inv_scale_y = (double)drows / rows;
softdouble scale_x = softdouble::one() / softdouble(inv_scale_x);
softdouble scale_y = softdouble::one() / softdouble(inv_scale_y);
Mat src(rows, cols, type), refdst(drows, dcols, type), dst;
RNG rnd(0x123456789abcdefULL);
for (int j = 0; j < rows; j++)
{
uint8_t* line = src.ptr(j);
for (int i = 0; i < cols; i++)
for (int c = 0; c < cn; c++)
{
double val = j < rows / 2 ? ( i < cols / 2 ? ((sin((i + 1)*CV_PI / 256.)*sin((j + 1)*CV_PI / 256.)*sin((cn + 4)*CV_PI / 8.) + 1.)*128.) :
(((i / 128 + j / 128) % 2) * 250 + (j / 128) % 2) ) :
( i < cols / 2 ? ((i / 128) * (85 - j / 256 * 40) * ((j / 128) % 2) + (7 - i / 128) * (85 - j / 256 * 40) * ((j / 128 + 1) % 2)) :
((uchar)rnd) ) ;
if (depth == CV_8U)
line[i*cn + c] = (uint8_t)val;
else if (depth == CV_16U)
((uint16_t*)line)[i*cn + c] = (uint16_t)val;
else if (depth == CV_16S)
((int16_t*)line)[i*cn + c] = (int16_t)val;
else if (depth == CV_32S)
((int32_t*)line)[i*cn + c] = (int32_t)val;
else
CV_Assert(0);
}
}
for (int j = 0; j < drows; j++)
{
softdouble src_row_flt = scale_y*(softdouble(j) + softdouble(0.5)) - softdouble(0.5);
int src_row = cvFloor(src_row_flt);
int64_t ycoeff1 = cvRound64((src_row_flt - softdouble(src_row))*softdouble(fixedOne));
int64_t ycoeff0 = fixedOne - ycoeff1;
for (int i = 0; i < dcols; i++)
{
softdouble src_col_flt = scale_x*(softdouble(i) + softdouble(0.5)) - softdouble(0.5);
int src_col = cvFloor(src_col_flt);
int64_t xcoeff1 = cvRound64((src_col_flt - softdouble(src_col))*softdouble(fixedOne));
int64_t xcoeff0 = fixedOne - xcoeff1;
uint8_t* dst_pt = refdst.ptr(j, i);
uint8_t* src_pt00 = src.ptr( src_row < 0 ? 0 : src_row >= rows ? rows - 1 : src_row ,
src_col < 0 ? 0 : src_col >= cols ? cols - 1 : src_col );
uint8_t* src_pt01 = src.ptr( src_row < 0 ? 0 : src_row >= rows ? rows - 1 : src_row ,
(src_col + 1) < 0 ? 0 : (src_col + 1) >= cols ? cols - 1 : (src_col + 1));
uint8_t* src_pt10 = src.ptr((src_row + 1) < 0 ? 0 : (src_row + 1) >= rows ? rows - 1 : (src_row + 1),
src_col < 0 ? 0 : src_col >= cols ? cols - 1 : src_col );
uint8_t* src_pt11 = src.ptr((src_row + 1) < 0 ? 0 : (src_row + 1) >= rows ? rows - 1 : (src_row + 1),
(src_col + 1) < 0 ? 0 : (src_col + 1) >= cols ? cols - 1 : (src_col + 1));
for (int c = 0; c < cn; c++)
{
if (depth == CV_8U)
eval4< uint8_t, fixedShiftU8>(xcoeff0, xcoeff1, ycoeff0, ycoeff1, c, src_pt00, src_pt01, src_pt10, src_pt11, dst_pt);
else if (depth == CV_16U)
eval4<uint16_t, fixedShiftU8>(xcoeff0, xcoeff1, ycoeff0, ycoeff1, c, src_pt00, src_pt01, src_pt10, src_pt11, dst_pt);
else if (depth == CV_16S)
eval4< int16_t, fixedShiftU8>(xcoeff0, xcoeff1, ycoeff0, ycoeff1, c, src_pt00, src_pt01, src_pt10, src_pt11, dst_pt);
else if (depth == CV_32S)
eval4< int32_t, fixedShiftU8>(xcoeff0, xcoeff1, ycoeff0, ycoeff1, c, src_pt00, src_pt01, src_pt10, src_pt11, dst_pt);
else
CV_Assert(0);
}
}
}
cv::resize(src, dst, Size(dcols, drows), 0, 0, cv::INTER_LINEAR_EXACT);
EXPECT_GE(0, cvtest::norm(refdst, dst, cv::NORM_L1))
<< "Resize " << cn << "-chan mat from " << cols << "x" << rows << " to " << dcols << "x" << drows << " failed with max diff " << cvtest::norm(refdst, dst, cv::NORM_INF);
}
}
PARAM_TEST_CASE(Resize_Bitexact, int)
{
public:
int depth;
virtual void SetUp()
{
depth = GET_PARAM(0);
}
double CountDiff(const Mat& src)
{
Mat dstExact; cv::resize(src, dstExact, Size(), 2, 1, INTER_NEAREST_EXACT);
Mat dstNonExact; cv::resize(src, dstNonExact, Size(), 2, 1, INTER_NEAREST);
return cv::norm(dstExact, dstNonExact, NORM_INF);
}
};
TEST_P(Resize_Bitexact, Nearest8U_vsNonExact)
{
Mat mat_color, mat_gray;
Mat src_color = imread(cvtest::findDataFile("shared/lena.png"));
Mat src_gray; cv::cvtColor(src_color, src_gray, COLOR_BGR2GRAY);
src_color.convertTo(mat_color, depth);
src_gray.convertTo(mat_gray, depth);
EXPECT_EQ(CountDiff(mat_color), 0) << "color, type: " << depth;
EXPECT_EQ(CountDiff(mat_gray), 0) << "gray, type: " << depth;
}
// Now INTER_NEAREST's convention and INTER_NEAREST_EXACT's one are different.
INSTANTIATE_TEST_CASE_P(DISABLED_Imgproc, Resize_Bitexact,
testing::Values(CV_8U, CV_16U, CV_32F, CV_64F)
);
TEST(Resize_Bitexact, Nearest8U)
{
Mat src[6], dst[6];
// 2x decimation
src[0] = (Mat_<uint8_t>(1, 6) << 0, 1, 2, 3, 4, 5);
dst[0] = (Mat_<uint8_t>(1, 3) << 1, 3, 5);
// decimation odd to 1
src[1] = (Mat_<uint8_t>(1, 5) << 0, 1, 2, 3, 4);
dst[1] = (Mat_<uint8_t>(1, 1) << 2);
// decimation n*2-1 to n
src[2] = (Mat_<uint8_t>(1, 5) << 0, 1, 2, 3, 4);
dst[2] = (Mat_<uint8_t>(1, 3) << 0, 2, 4);
// decimation n*2+1 to n
src[3] = (Mat_<uint8_t>(1, 5) << 0, 1, 2, 3, 4);
dst[3] = (Mat_<uint8_t>(1, 2) << 1, 3);
// zoom
src[4] = (Mat_<uint8_t>(3, 5) <<
0, 1, 2, 3, 4,
5, 6, 7, 8, 9,
10, 11, 12, 13, 14);
dst[4] = (Mat_<uint8_t>(5, 7) <<
0, 1, 1, 2, 3, 3, 4,
0, 1, 1, 2, 3, 3, 4,
5, 6, 6, 7, 8, 8, 9,
10, 11, 11, 12, 13, 13, 14,
10, 11, 11, 12, 13, 13, 14);
src[5] = (Mat_<uint8_t>(2, 3) <<
0, 1, 2,
3, 4, 5);
dst[5] = (Mat_<uint8_t>(4, 6) <<
0, 0, 1, 1, 2, 2,
0, 0, 1, 1, 2, 2,
3, 3, 4, 4, 5, 5,
3, 3, 4, 4, 5, 5);
for (int i = 0; i < 6; i++)
{
Mat calc;
resize(src[i], calc, dst[i].size(), 0, 0, INTER_NEAREST_EXACT);
EXPECT_EQ(cvtest::norm(calc, dst[i], cv::NORM_L1), 0);
resize(src[i].t(), calc, dst[i].t().size(), 0, 0, INTER_NEAREST_EXACT);
EXPECT_EQ(cvtest::norm(calc, dst[i].t(), cv::NORM_L1), 0);
}
}
// Regression test for #28429: INTER_NEAREST_EXACT matches Pillow's
// center-of-pixel nearest-neighbor mapping.
TEST(Resize_Bitexact, NearestExact_PillowCompat)
{
auto center_pixel_map = [](int src_dim, int dst_dim, std::vector<int>& mapping) {
softdouble scale = softdouble(src_dim) / softdouble(dst_dim);
softdouble f = scale * softdouble(0.5);
mapping.resize(dst_dim);
for (int i = 0; i < dst_dim; i++)
{
mapping[i] = std::min(cvTrunc(f), src_dim - 1);
f += scale;
}
};
// Test dimension pairs including multiples of 64 that triggered the bug
const int cases[][4] = {
{128, 147, 160, 160}, // original reproducer from #28429
{128, 128, 160, 160}, // square with problematic height
{192, 192, 256, 256}, // another multiple of 64
{129, 147, 160, 160}, // non-problematic control case
};
for (const auto& c : cases)
{
int src_h = c[0], src_w = c[1], dst_h = c[2], dst_w = c[3];
std::vector<int> x_map, y_map;
center_pixel_map(src_w, dst_w, x_map);
center_pixel_map(src_h, dst_h, y_map);
Mat src(src_h, src_w, CV_8UC3);
randu(src, Scalar::all(0), Scalar::all(256));
Mat result;
resize(src, result, Size(dst_w, dst_h), 0, 0, INTER_NEAREST_EXACT);
int nerrs = 0;
for (int y = 0; y < dst_h; y++)
{
for (int x = 0; x < dst_w; x++)
{
Vec3b expected = src.at<Vec3b>(y_map[y], x_map[x]);
Vec3b actual = result.at<Vec3b>(y, x);
nerrs += expected != actual;
if (nerrs <= 10) {
EXPECT_EQ(expected, actual)
<< "Mismatch at dst(" << y << "," << x << ") -> src("
<< y_map[y] << "," << x_map[x] << ") for "
<< src_h << "x" << src_w << " -> " << dst_h << "x" << dst_w;
}
}
}
ASSERT_EQ(nerrs, 0) << "nerrs=" << nerrs << " when"
<< " src_h=" << src_h << ", src_w=" << src_w
<< " dst_h=" << dst_h << ", dst_w=" << dst_w;
}
}
}} // namespace
View Git Blame Copy Permalink