Skip to content

O
opencv
Commit f6c57388 authored by Alexander Alekhin's avatar Alexander Alekhin
Browse files
Options

Merge remote-tracking branch 'upstream/3.4' into merge-3.4

parents f663e8f9 054c7962
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 275 additions and 63 deletions
modules/dnn/perf/perf_convolution3d.cpp 0 → 100644
View file @ f6c57388
// 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 "perf_precomp.hpp"
#include <opencv2/dnn/shape_utils.hpp>
namespace opencv_test {
struct Conv3DParam_t {
int kernel[3];
struct BlobShape { int dims[5]; } shapeIn;
int outCN;
int groups;
int stride[3];
int dilation[3];
int pad[6];
const char* padMode;
bool hasBias;
double declared_flops;
};
// Details: #12142
static const Conv3DParam_t testConvolution3DConfigs[] = {
{{3, 3, 3}, {{1, 6, 10, 38, 50}}, 6, 1, {1, 1, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "VALID", true, 26956800.},
{{3, 3, 3}, {{1, 2, 19, 19, 19}}, 2, 2, {2, 2, 2}, {1, 1, 1}, {1, 1, 1, 1, 1, 1}, "", true, 218000.},
{{3, 3, 3}, {{1, 2, 25, 19, 19}}, 2, 2, {1, 2, 2}, {1, 1, 1}, {2, 2, 2, 2, 2, 2}, "SAME", false, 545000.},
{{3, 3, 3}, {{1, 11, 9, 150, 200}}, 11, 1, {1, 1, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "VALID", true, 1342562760.},
{{3, 3, 3}, {{1, 10, 98, 10, 10}}, 10, 1, {1, 1, 1}, {1, 1, 1}, {1, 0, 1, 1, 0,1}, "SAME", false, 53018000.},
{{5, 5, 5}, {{1, 6, 19, 19, 19}}, 6, 2, {1, 1, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "", false, 30395250.},
{{5, 5, 5}, {{1, 4, 50, 19, 19}}, 4, 1, {2, 2, 2}, {1, 1, 1}, {1, 1, 1, 1, 1, 1}, "VALID", false, 5893888.},
{{5, 5, 5}, {{1, 3, 75, 75, 100}}, 3, 1, {1, 1, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "SAME", true, 1267312500.},
{{5, 5, 5}, {{1, 2, 21, 75, 100}}, 2, 1, {1, 1, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "", true, 116103744.},
{{5, 5, 5}, {{1, 4, 40, 75, 75}}, 4, 1, {2, 2, 2}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "", false, 93405312.},
{{7, 7, 7}, {{1, 6, 15, 19, 19}}, 6, 1, {2, 1, 1}, {1, 1, 1}, {3, 3, 3, 3, 3, 3}, "SAME", true, 71339376.},
{{7, 7, 7}, {{1, 2, 38, 38, 38}}, 2, 1, {1, 2, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "", false, 44990464.},
{{1, 1, 1}, {{1, 4, 9, 10, 10}}, 4, 1, {1, 1, 2}, {1, 1, 1}, {1, 1, 1, 1, 1, 1}, "VALID", false, 16200.},
{{3, 1, 4}, {{1, 14, 5, 10, 10}}, 14, 1, {1, 1, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "SAME", false, 2359000.},
{{1, 1, 1}, {{1, 8, 1, 10, 10}}, 8, 8, {1, 1, 1}, {1, 1, 1}, {1, 1, 1, 1, 1, 1}, "", true, 58752.},
{{3, 4, 2}, {{1, 4, 8, 10, 10}}, 4, 4, {1, 2, 1}, {1, 1, 1}, {0, 0, 0, 0, 0, 0}, "", true, 166752.}
};
struct Conv3DParamID
{
enum {
CONV_0 = 0,
CONV_100 = 16,
CONV_LAST = sizeof(testConvolution3DConfigs) / sizeof(testConvolution3DConfigs[0])
};
int val_; \
Conv3DParamID(int val = 0) : val_(val) {}
operator int() const { return val_; }
static ::testing::internal::ParamGenerator<Conv3DParamID> all()
{
#if 0
enum { NUM = (int)CONV_LAST };
#else
enum { NUM = (int)CONV_100 };
#endif
Conv3DParamID v_[NUM]; for (int i = 0; i < NUM; ++i) { v_[i] = Conv3DParamID(i); } // reduce generated code size
return ::testing::ValuesIn(v_, v_ + NUM);
}
}; \
static inline void PrintTo(const Conv3DParamID& v, std::ostream* os)
{
CV_Assert((int)v >= 0); CV_Assert((int)v < Conv3DParamID::CONV_LAST);
const Conv3DParam_t& p = testConvolution3DConfigs[(int)v];
*os << "GFLOPS=" << cv::format("%.3f", p.declared_flops * 1e-9)
<< ", K=[" << p.kernel[0] << " x " << p.kernel[1] << " x " << p.kernel[2] << "]"
<< ", IN={" << p.shapeIn.dims[0] << ", " << p.shapeIn.dims[1] << ", " << p.shapeIn.dims[2] << ", " << p.shapeIn.dims[3] << ", " << p.shapeIn.dims[4] << "}"
<< ", OCN=" << p.outCN;
if (p.groups > 1)
*os << ", G=" << p.groups;
if (p.stride[0] * p.stride[1] * p.stride[2] != 1)
*os << ", S=[" << p.stride[0] << " x " << p.stride[1] << " x " << p.stride[2] << "]";
if (p.dilation[0] * p.dilation[1] * p.dilation[2] != 1)
*os << ", D=[" << p.dilation[0] << " x " << p.dilation[1] << " x " << p.dilation[2] << "]";
if (p.pad[0] != 0 && p.pad[1] != 0 && p.pad[2] != 0 &&
p.pad[3] != 0 && p.pad[4] != 0 && p.pad[5] != 0)
*os << ", P=(" << p.pad[0] << ", " << p.pad[3] << ") x ("
<< p.pad[1] << ", " << p.pad[4] << ") x ("
<< p.pad[2] << ", " << p.pad[5] << ")";
if (!((std::string)p.padMode).empty())
*os << ", PM=" << ((std::string)p.padMode);
if (p.hasBias)
*os << ", BIAS";
}
typedef tuple<Conv3DParamID, tuple<Backend, Target> > Conv3DTestParam_t;
typedef TestBaseWithParam<Conv3DTestParam_t> Conv3D;
PERF_TEST_P_(Conv3D, conv3d)
{
int test_id = (int)get<0>(GetParam());
ASSERT_GE(test_id, 0); ASSERT_LT(test_id, Conv3DParamID::CONV_LAST);
const Conv3DParam_t& params = testConvolution3DConfigs[test_id];
double declared_flops = params.declared_flops;
DictValue kernel = DictValue::arrayInt(&params.kernel[0], 3);
DictValue stride = DictValue::arrayInt(&params.stride[0], 3);
DictValue pad = DictValue::arrayInt(&params.pad[0], 6);
DictValue dilation = DictValue::arrayInt(&params.dilation[0], 3);
MatShape inputShape = MatShape(params.shapeIn.dims, params.shapeIn.dims + 5);
int outChannels = params.outCN;
int groups = params.groups;
std::string padMode(params.padMode);
bool hasBias = params.hasBias;
Backend backendId = get<0>(get<1>(GetParam()));
Target targetId = get<1>(get<1>(GetParam()));
if (targetId != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
int inChannels = inputShape[1];
int sz[] = {outChannels, inChannels / groups, params.kernel[0], params.kernel[1], params.kernel[2]};
Mat weights(5, &sz[0], CV_32F);
randu(weights, -1.0f, 1.0f);
LayerParams lp;
lp.set("kernel_size", kernel);
lp.set("pad", pad);
if (!padMode.empty())
lp.set("pad_mode", padMode);
lp.set("stride", stride);
lp.set("dilation", dilation);
lp.set("num_output", outChannels);
lp.set("group", groups);
lp.set("bias_term", hasBias);
lp.type = "Convolution";
lp.name = "testLayer";
lp.blobs.push_back(weights);
if (hasBias)
{
Mat bias(1, outChannels, CV_32F);
randu(bias, -1.0f, 1.0f);
lp.blobs.push_back(bias);
}
int inpSz[] = {1, inChannels, inputShape[2], inputShape[3], inputShape[4]};
Mat input(5, &inpSz[0], CV_32F);
randu(input, -1.0f, 1.0f);
Net net;
net.addLayerToPrev(lp.name, lp.type, lp);
net.setInput(input);
net.setPreferableBackend(backendId);
net.setPreferableTarget(targetId);
Mat output = net.forward();
MatShape netInputShape = shape(input);
size_t weightsMemory = 0, blobsMemory = 0;
net.getMemoryConsumption(netInputShape, weightsMemory, blobsMemory);
int64 flops = net.getFLOPS(netInputShape);
CV_Assert(flops > 0);
std::cout
<< "IN=" << divUp(input.total() * input.elemSize(), 1u<<10) << " Kb " << netInputShape
<< " OUT=" << divUp(output.total() * output.elemSize(), 1u<<10) << " Kb " << shape(output)
<< " Weights(parameters): " << divUp(weightsMemory, 1u<<10) << " Kb"
<< " MFLOPS=" << flops * 1e-6 << std::endl;
TEST_CYCLE()
{
Mat res = net.forward();
}
EXPECT_NEAR(flops, declared_flops, declared_flops * 1e-6);
SANITY_CHECK_NOTHING();
}
INSTANTIATE_TEST_CASE_P(/**/, Conv3D, Combine(
Conv3DParamID::all(),
dnnBackendsAndTargets(false, false) // defined in ../test/test_common.hpp
));
} // namespace
modules/dnn/src/layers/layers_common.cpp
View file @ f6c57388
......@@ -175,11 +175,13 @@ void getPoolingKernelParams(const LayerParams &params, std::vector<size_t>& kern
}
void getConvolutionKernelParams(const LayerParams &params, std::vector<size_t>& kernel, std::vector<size_t>& pads_begin,
std::vector<size_t>& pads_end, std::vector<size_t>& strides, std::vector<size_t>& dilations, cv::String &padMode)
std::vector<size_t>& pads_end, std::vector<size_t>& strides,
std::vector<size_t>& dilations, cv::String &padMode, std::vector<size_t>& adjust_pads)
{
util::getKernelSize(params, kernel);
util::getStrideAndPadding(params, pads_begin, pads_end, strides, padMode, kernel.size());
util::getParameter(params, "dilation", "dilation", dilations, true, std::vector<size_t>(kernel.size(), 1));
util::getParameter(params, "adj", "adj", adjust_pads, true, std::vector<size_t>(kernel.size(), 0));
for (int i = 0; i < dilations.size(); i++)
CV_Assert(dilations[i] > 0);
......
modules/dnn/src/layers/layers_common.hpp
View file @ f6c57388
......@@ -60,7 +60,8 @@ namespace cv
namespace dnn
{
void getConvolutionKernelParams(const LayerParams &params, std::vector<size_t>& kernel, std::vector<size_t>& pads_begin,
std::vector<size_t>& pads_end, std::vector<size_t>& strides, std::vector<size_t>& dilations, cv::String &padMode);
std::vector<size_t>& pads_end, std::vector<size_t>& strides, std::vector<size_t>& dilations,
cv::String &padMode, std::vector<size_t>& adjust_pads);
void getPoolingKernelParams(const LayerParams &params, std::vector<size_t>& kernel, bool &globalPooling,
std::vector<size_t>& pads_begin, std::vector<size_t>& pads_end, std::vector<size_t>& strides, cv::String &padMode);
......
modules/dnn/src/layers/pooling_layer.cpp
View file @ f6c57388
This diff is collapsed.
modules/dnn/src/onnx/onnx_importer.cpp
View file @ f6c57388
......@@ -682,42 +682,37 @@ void ONNXImporter::populateNet(Net dstNet)
layerParams.set("num_output", layerParams.blobs[0].size[1] * layerParams.get<int>("group", 1));
layerParams.set("bias_term", node_proto.input_size() == 3);
if (!layerParams.has("kernel_size"))
CV_Error(Error::StsNotImplemented,
"Required attribute 'kernel_size' is not present.");
if (layerParams.has("output_shape"))
{
const DictValue& outShape = layerParams.get("output_shape");
DictValue strides = layerParams.get("stride");
DictValue kernel = layerParams.get("kernel_size");
if (outShape.size() != 4)
CV_Error(Error::StsNotImplemented, "Output shape must have 4 elements.");
DictValue stride = layerParams.get("stride");
const int strideY = stride.getIntValue(0);
const int strideX = stride.getIntValue(1);
const int outH = outShape.getIntValue(2);
const int outW = outShape.getIntValue(3);
if (layerParams.get<String>("pad_mode") == "SAME")
String padMode;
std::vector<int> adjust_pads;
if (layerParams.has("pad_mode"))
{
layerParams.set("adj_w", (outW - 1) % strideX);
layerParams.set("adj_h", (outH - 1) % strideY);
}
else if (layerParams.get<String>("pad_mode") == "VALID")
{
if (!layerParams.has("kernel_size"))
CV_Error(Error::StsNotImplemented,
"Required attribute 'kernel_size' is not present.");
DictValue kernel = layerParams.get("kernel_size");
layerParams.set("adj_h", (outH - kernel.getIntValue(0)) % strideY);
layerParams.set("adj_w", (outW - kernel.getIntValue(1)) % strideX);
padMode = toUpperCase(layerParams.get<String>("pad_mode"));
if (padMode != "SAME" && padMode != "VALID")
CV_Error(Error::StsError, "Unsupported padding mode " + padMode);
for (int i = 0; i < strides.size(); i++)
{
int sz = outShape.get<int>(2 + i);
int stride = strides.get<int>(i);
adjust_pads.push_back(padMode == "SAME"? (sz - 1) % stride :
(sz - kernel.get<int>(i)) % stride);
}
layerParams.set("adj", DictValue::arrayInt(&adjust_pads[0], adjust_pads.size()));
}
}
else if (layerParams.has("output_padding"))
{
const DictValue& adj_pad = layerParams.get("output_padding");
if (adj_pad.size() != 2)
CV_Error(Error::StsNotImplemented, "Deconvolution3D layer is not supported");
layerParams.set("adj_w", adj_pad.get<int>(1));
layerParams.set("adj_h", adj_pad.get<int>(0));
replaceLayerParam(layerParams, "output_padding", "adj");
}
}
else if (layer_type == "Transpose")
......
modules/dnn/test/test_onnx_importer.cpp
View file @ f6c57388
......@@ -100,8 +100,8 @@ TEST_P(Test_ONNX_layers, Convolution3D)
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
testONNXModels("conv3d");
testONNXModels("conv3d_bias");
}
......@@ -127,6 +127,19 @@ TEST_P(Test_ONNX_layers, Deconvolution)
testONNXModels("deconv_adjpad_2d", npy, 0, 0, false, false);
}
TEST_P(Test_ONNX_layers, Deconvolution3D)
{
#if defined(INF_ENGINE_RELEASE)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_2018R5);
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
testONNXModels("deconv3d");
testONNXModels("deconv3d_bias");
testONNXModels("deconv3d_pad");
testONNXModels("deconv3d_adjpad");
}
TEST_P(Test_ONNX_layers, Dropout)
{
testONNXModels("dropout");
......@@ -185,8 +198,8 @@ TEST_P(Test_ONNX_layers, MaxPooling3D)
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
testONNXModels("max_pool3d");
}
......@@ -195,11 +208,21 @@ TEST_P(Test_ONNX_layers, AvePooling3D)
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
testONNXModels("ave_pool3d");
}
TEST_P(Test_ONNX_layers, PoolConv3D)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
testONNXModels("pool_conv_3d");
}
TEST_P(Test_ONNX_layers, BatchNormalization)
{
testONNXModels("batch_norm");
......@@ -579,10 +602,10 @@ TEST_P(Test_ONNX_nets, Resnet34_kinetics)
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
String onnxmodel = findDataFile("dnn/resnet-34_kinetics.onnx", false);
String onnxmodel = findDataFile("dnn/resnet-34_kinetics.onnx");
Mat image0 = imread(findDataFile("dnn/dog416.png"));
Mat image1 = imread(findDataFile("dnn/street.png"));
......
modules/dnn/test/test_tf_importer.cpp
View file @ f6c57388
......@@ -136,8 +136,8 @@ TEST_P(Test_TensorFlow_layers, Convolution3D)
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
runTensorFlowNet("conv3d");
}
......@@ -243,8 +243,8 @@ TEST_P(Test_TensorFlow_layers, MaxPooling3D)
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
runTensorFlowNet("max_pool3d");
}
......@@ -253,8 +253,8 @@ TEST_P(Test_TensorFlow_layers, AvePooling3D)
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2019010000)
throw SkipTestException("Test is enabled starts from 2019R1");
#endif
if (backend != DNN_BACKEND_INFERENCE_ENGINE || target != DNN_TARGET_CPU)
throw SkipTestException("Only DLIE backend on CPU is supported");
if (target != DNN_TARGET_CPU)
throw SkipTestException("Only CPU is supported");
runTensorFlowNet("ave_pool3d");
}
......
modules/imgproc/src/filter.simd.hpp
View file @ f6c57388
......@@ -84,6 +84,7 @@ Ptr<BaseFilter> getLinearFilter(
#ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
typedef int CV_DECL_ALIGNED(1) unaligned_int;
#define VEC_ALIGN CV_MALLOC_ALIGN
int FilterEngine__start(FilterEngine& this_, const Size &_wholeSize, const Size &sz, const Point &ofs)
......@@ -1049,7 +1050,7 @@ struct SymmColumnVec_32s8u
s0 = v_muladd(v_cvt_f32(v_load(src[k] + i) + v_load(src[-k] + i)), v_setall_f32(ky[k]), s0);
v_int32x4 s32 = v_round(s0);
v_int16x8 s16 = v_pack(s32, s32);
*(int*)(dst + i) = v_reinterpret_as_s32(v_pack_u(s16, s16)).get0();
*(unaligned_int*)(dst + i) = v_reinterpret_as_s32(v_pack_u(s16, s16)).get0();
i += v_int32x4::nlanes;
}
}
......@@ -1104,7 +1105,7 @@ struct SymmColumnVec_32s8u
s0 = v_muladd(v_cvt_f32(v_load(src[k] + i) - v_load(src[-k] + i)), v_setall_f32(ky[k]), s0);
v_int32x4 s32 = v_round(s0);
v_int16x8 s16 = v_pack(s32, s32);
*(int*)(dst + i) = v_reinterpret_as_s32(v_pack_u(s16, s16)).get0();
*(unaligned_int*)(dst + i) = v_reinterpret_as_s32(v_pack_u(s16, s16)).get0();
i += v_int32x4::nlanes;
}
}
......@@ -2129,7 +2130,7 @@ struct FilterVec_8u
s0 = v_muladd(v_cvt_f32(v_reinterpret_as_s32(v_load_expand_q(src[k] + i))), v_setall_f32(kf[k]), s0);
v_int32x4 s32 = v_round(s0);
v_int16x8 s16 = v_pack(s32, s32);
*(int*)(dst + i) = v_reinterpret_as_s32(v_pack_u(s16, s16)).get0();
*(unaligned_int*)(dst + i) = v_reinterpret_as_s32(v_pack_u(s16, s16)).get0();
i += v_int32x4::nlanes;
}
return i;
......
modules/imgproc/src/smooth.simd.hpp
View file @ f6c57388
......@@ -334,7 +334,7 @@ void hlineSmooth3Naba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, const
{
int src_idx = borderInterpolate(-1, len, borderType);
for (int k = 0; k < cn; k++)
((uint16_t*)dst)[k] = ((uint16_t*)m)[1] * src[k] + ((uint16_t*)m)[0] * ((uint16_t)(src[cn + k]) + (uint16_t)(src[src_idx*cn + k]));
((uint16_t*)dst)[k] = saturate_cast<uint16_t>(((uint16_t*)m)[1] * (uint32_t)(src[k]) + ((uint16_t*)m)[0] * ((uint32_t)(src[cn + k]) + (uint32_t)(src[src_idx*cn + k])));
}
else
{
......@@ -354,14 +354,14 @@ void hlineSmooth3Naba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, const
v_mul_wrap(vx_load_expand(src), v_mul1));
#endif
for (; i < lencn; i++, src++, dst++)
*((uint16_t*)dst) = ((uint16_t*)m)[1] * src[0] + ((uint16_t*)m)[0] * ((uint16_t)(src[-cn]) + (uint16_t)(src[cn]));
*((uint16_t*)dst) = saturate_cast<uint16_t>(((uint16_t*)m)[1] * (uint32_t)(src[0]) + ((uint16_t*)m)[0] * ((uint32_t)(src[-cn]) + (uint32_t)(src[cn])));
// Point that fall right from border
if (borderType != BORDER_CONSTANT)// If BORDER_CONSTANT out of border values are equal to zero and could be skipped
{
int src_idx = (borderInterpolate(len, len, borderType) - (len - 1))*cn;
for (int k = 0; k < cn; k++)
((uint16_t*)dst)[k] = ((uint16_t*)m)[1] * src[k] + ((uint16_t*)m)[0] * ((uint16_t)(src[k - cn]) + (uint16_t)(src[src_idx + k]));
((uint16_t*)dst)[k] = saturate_cast<uint16_t>(((uint16_t*)m)[1] * (uint32_t)(src[k]) + ((uint16_t*)m)[0] * ((uint32_t)(src[k - cn]) + (uint32_t)(src[src_idx + k])));
}
else
{
......@@ -896,8 +896,8 @@ void hlineSmooth5Nabcba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, cons
int idxp2 = borderInterpolate(3, len, borderType)*cn;
for (int k = 0; k < cn; k++)
{
((uint16_t*)dst)[k] = ((uint16_t*)m)[1] * ((uint16_t)(src[k + idxm1]) + (uint16_t)(src[k + cn])) + ((uint16_t*)m)[2] * src[k] + ((uint16_t*)m)[0] * ((uint16_t)(src[k + idxp1]) + (uint16_t)(src[k + idxm2]));
((uint16_t*)dst)[k + cn] = ((uint16_t*)m)[0] * ((uint16_t)(src[k + idxm1]) + (uint16_t)(src[k + idxp2])) + ((uint16_t*)m)[1] * ((uint16_t)(src[k]) + (uint16_t)(src[k + idxp1])) + ((uint16_t*)m)[2] * src[k + cn];
((uint16_t*)dst)[k] = saturate_cast<uint16_t>(((uint16_t*)m)[1] * ((uint32_t)(src[k + idxm1]) + (uint32_t)(src[k + cn])) + ((uint16_t*)m)[2] * (uint32_t)(src[k]) + ((uint16_t*)m)[0] * ((uint32_t)(src[k + idxp1]) + (uint32_t)(src[k + idxm2])));
((uint16_t*)dst)[k + cn] = saturate_cast<uint16_t>(((uint16_t*)m)[0] * ((uint32_t)(src[k + idxm1]) + (uint32_t)(src[k + idxp2])) + ((uint16_t*)m)[1] * ((uint32_t)(src[k]) + (uint32_t)(src[k + idxp1])) + ((uint16_t*)m)[2] * (uint32_t)(src[k + cn]));
}
}
}
......@@ -907,7 +907,7 @@ void hlineSmooth5Nabcba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, cons
for (int k = 0; k < cn; k++)
{
dst[k] = m[2] * src[k] + m[1] * src[k + cn] + m[0] * src[k + 2 * cn];
((uint16_t*)dst)[k + cn] = ((uint16_t*)m)[1] * ((uint16_t)(src[k]) + (uint16_t)(src[k + 2 * cn])) + ((uint16_t*)m)[2] * src[k + cn];
((uint16_t*)dst)[k + cn] = saturate_cast<uint16_t>(((uint16_t*)m)[1] * ((uint32_t)(src[k]) + (uint32_t)(src[k + 2 * cn])) + ((uint16_t*)m)[2] * (uint32_t)(src[k + cn]));
dst[k + 2 * cn] = m[0] * src[k] + m[1] * src[k + cn] + m[2] * src[k + 2 * cn];
}
else
......@@ -918,9 +918,9 @@ void hlineSmooth5Nabcba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, cons
int idxp2 = borderInterpolate(4, len, borderType)*cn;
for (int k = 0; k < cn; k++)
{
((uint16_t*)dst)[k] = ((uint16_t*)m)[2] * src[k] + ((uint16_t*)m)[1] * ((uint16_t)(src[k + cn]) + (uint16_t)(src[k + idxm1])) + ((uint16_t*)m)[0] * ((uint16_t)(src[k + 2 * cn]) + (uint16_t)(src[k + idxm2]));
((uint16_t*)dst)[k + cn] = ((uint16_t*)m)[2] * src[k + cn] + ((uint16_t*)m)[1] * ((uint16_t)(src[k]) + (uint16_t)(src[k + 2 * cn])) + ((uint16_t*)m)[0] * ((uint16_t)(src[k + idxm1]) + (uint16_t)(src[k + idxp1]));
((uint16_t*)dst)[k + 2 * cn] = ((uint16_t*)m)[0] * ((uint16_t)(src[k]) + (uint16_t)(src[k + idxp2])) + ((uint16_t*)m)[1] * ((uint16_t)(src[k + cn]) + (uint16_t)(src[k + idxp1])) + ((uint16_t*)m)[2] * src[k + 2 * cn];
((uint16_t*)dst)[k] = saturate_cast<uint16_t>(((uint16_t*)m)[2] * (uint32_t)(src[k]) + ((uint16_t*)m)[1] * ((uint32_t)(src[k + cn]) + (uint32_t)(src[k + idxm1])) + ((uint16_t*)m)[0] * ((uint32_t)(src[k + 2 * cn]) + (uint32_t)(src[k + idxm2])));
((uint16_t*)dst)[k + cn] = saturate_cast<uint16_t>(((uint16_t*)m)[2] * (uint32_t)(src[k + cn]) + ((uint16_t*)m)[1] * ((uint32_t)(src[k]) + (uint32_t)(src[k + 2 * cn])) + ((uint16_t*)m)[0] * ((uint32_t)(src[k + idxm1]) + (uint32_t)(src[k + idxp1])));
((uint16_t*)dst)[k + 2 * cn] = saturate_cast<uint16_t>(((uint16_t*)m)[0] * ((uint32_t)(src[k]) + (uint32_t)(src[k + idxp2])) + ((uint16_t*)m)[1] * ((uint32_t)(src[k + cn]) + (uint32_t)(src[k + idxp1])) + ((uint16_t*)m)[2] * (uint32_t)(src[k + 2 * cn]));
}
}
}
......@@ -933,8 +933,8 @@ void hlineSmooth5Nabcba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, cons
int idxm1 = borderInterpolate(-1, len, borderType)*cn;
for (int k = 0; k < cn; k++)
{
((uint16_t*)dst)[k] = ((uint16_t*)m)[2] * src[k] + ((uint16_t*)m)[1] * ((uint16_t)(src[cn + k]) + (uint16_t)(src[idxm1 + k])) + ((uint16_t*)m)[0] * ((uint16_t)(src[2 * cn + k]) + (uint16_t)(src[idxm2 + k]));
((uint16_t*)dst)[k + cn] = ((uint16_t*)m)[1] * ((uint16_t)(src[k]) + (uint16_t)(src[2 * cn + k])) + ((uint16_t*)m)[2] * src[cn + k] + ((uint16_t*)m)[0] * ((uint16_t)(src[3 * cn + k]) + (uint16_t)(src[idxm1 + k]));
((uint16_t*)dst)[k] = saturate_cast<uint16_t>(((uint16_t*)m)[2] * (uint32_t)(src[k]) + ((uint16_t*)m)[1] * ((uint32_t)(src[cn + k]) + (uint32_t)(src[idxm1 + k])) + ((uint16_t*)m)[0] * ((uint32_t)(src[2 * cn + k]) + (uint32_t)(src[idxm2 + k])));
((uint16_t*)dst)[k + cn] = saturate_cast<uint16_t>(((uint16_t*)m)[1] * ((uint32_t)(src[k]) + (uint32_t)(src[2 * cn + k])) + ((uint16_t*)m)[2] * (uint32_t)(src[cn + k]) + ((uint16_t*)m)[0] * ((uint32_t)(src[3 * cn + k]) + (uint32_t)(src[idxm1 + k])));
}
}
else
......@@ -942,7 +942,7 @@ void hlineSmooth5Nabcba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, cons
for (int k = 0; k < cn; k++)
{
dst[k] = m[2] * src[k] + m[1] * src[cn + k] + m[0] * src[2 * cn + k];
((uint16_t*)dst)[k + cn] = ((uint16_t*)m)[1] * ((uint16_t)(src[k]) + (uint16_t)(src[2 * cn + k])) + ((uint16_t*)m)[2] * src[cn + k] + ((uint16_t*)m)[0] * src[3 * cn + k];
((uint16_t*)dst)[k + cn] = saturate_cast<uint16_t>(((uint16_t*)m)[1] * ((uint32_t)(src[k]) + (uint32_t)(src[2 * cn + k])) + ((uint16_t*)m)[2] * (uint32_t)(src[cn + k]) + ((uint16_t*)m)[0] * (uint32_t)(src[3 * cn + k]));
}
}
......@@ -960,7 +960,7 @@ void hlineSmooth5Nabcba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, cons
v_mul_wrap(vx_load_expand(src), v_mul2));
#endif
for (; i < lencn; i++, src++, dst++)
*((uint16_t*)dst) = ((uint16_t*)m)[0] * ((uint16_t)(src[-2 * cn]) + (uint16_t)(src[2 * cn])) + ((uint16_t*)m)[1] * ((uint16_t)(src[-cn]) + (uint16_t)(src[cn])) + ((uint16_t*)m)[2] * src[0];
*((uint16_t*)dst) = saturate_cast<uint16_t>(((uint16_t*)m)[0] * ((uint32_t)(src[-2 * cn]) + (uint32_t)(src[2 * cn])) + ((uint16_t*)m)[1] * ((uint32_t)(src[-cn]) + (uint32_t)(src[cn])) + ((uint16_t*)m)[2] * (uint32_t)(src[0]));
// Points that fall right from border
if (borderType != BORDER_CONSTANT)// If BORDER_CONSTANT out of border values are equal to zero and could be skipped
......@@ -969,15 +969,15 @@ void hlineSmooth5Nabcba<uint8_t, ufixedpoint16>(const uint8_t* src, int cn, cons
int idxp2 = (borderInterpolate(len + 1, len, borderType) - (len - 2))*cn;
for (int k = 0; k < cn; k++)
{
((uint16_t*)dst)[k] = ((uint16_t*)m)[0] * ((uint16_t)(src[k - 2 * cn]) + (uint16_t)(src[idxp1 + k])) + ((uint16_t*)m)[1] * ((uint16_t)(src[k - cn]) + (uint16_t)(src[k + cn])) + ((uint16_t*)m)[2] * src[k];
((uint16_t*)dst)[k + cn] = ((uint16_t*)m)[0] * ((uint16_t)(src[k - cn]) + (uint16_t)(src[idxp2 + k])) + ((uint16_t*)m)[1] * ((uint16_t)(src[k]) + (uint16_t)(src[idxp1 + k])) + ((uint16_t*)m)[2] * src[k + cn];
((uint16_t*)dst)[k] = saturate_cast<uint16_t>(((uint16_t*)m)[0] * ((uint32_t)(src[k - 2 * cn]) + (uint32_t)(src[idxp1 + k])) + ((uint16_t*)m)[1] * ((uint32_t)(src[k - cn]) + (uint32_t)(src[k + cn])) + ((uint16_t*)m)[2] * (uint32_t)(src[k]));
((uint16_t*)dst)[k + cn] = saturate_cast<uint16_t>(((uint16_t*)m)[0] * ((uint32_t)(src[k - cn]) + (uint32_t)(src[idxp2 + k])) + ((uint16_t*)m)[1] * ((uint32_t)(src[k]) + (uint32_t)(src[idxp1 + k])) + ((uint16_t*)m)[2] * (uint32_t)(src[k + cn]));
}
}
else
{
for (int k = 0; k < cn; k++)
{
((uint16_t*)dst)[k] = ((uint16_t*)m)[0] * src[k - 2 * cn] + ((uint16_t*)m)[1] * ((uint16_t)(src[k - cn]) + (uint16_t)(src[k + cn])) + ((uint16_t*)m)[2] * src[k];
((uint16_t*)dst)[k] = saturate_cast<uint16_t>(((uint16_t*)m)[0] * (uint32_t)(src[k - 2 * cn]) + ((uint16_t*)m)[1] * ((uint32_t)(src[k - cn]) + (uint32_t)(src[k + cn])) + ((uint16_t*)m)[2] * (uint32_t)(src[k]));
dst[k + cn] = m[0] * src[k - cn] + m[1] * src[k] + m[2] * src[k + cn];
}
}
......
modules/imgproc/test/test_smooth_bitexact.cpp
View file @ f6c57388
......@@ -158,4 +158,12 @@ TEST(GaussianBlur_Bitexact, Linear8U)
}
}
TEST(GaussianBlur_Bitexact, regression_15015)
{
Mat src(100,100,CV_8UC3,Scalar(255,255,255));
Mat dst;
GaussianBlur(src, dst, Size(5, 5), 9);
ASSERT_EQ(0.0, cvtest::norm(dst, src, NORM_INF));
}
}} // namespace
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment