ippiFilterGaussianBorder_8u_C1R - wrong results.

[EShamur]

Hi,

I am using IPP2018

I run Gaussian filter (ippiFilterGaussianBorder_8u_C1R)  , kernelSize = 3, sigma = 0.56, on single channel src image.

Src image size is 10x10. The image is flat ie all image data is set to 5.

I expect all dst image data to be 5 as well, However, I got strange result - dst image contains data = 4 and 3. This must be wrong.

Is it rounding issue? How would I work around it? Need your advise.

src
------
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5
5 5 5 5 5 5 5 5 5 5

dst
------
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3
4 4 4 4 4 4 4 4 3 3

Here is my code:

int main(int argc, _TCHAR* argv[])
{
int WIDTH = 10; /* image width */
int HEIGHT = 10; /* image height */
IppStatus status = ippStsNoErr;
Ipp8u* psrc=NULL, *pDst = NULL; /* Pointers to source/destination images */
int srcStep = 0, dstStep = 0; /* Steps, in bytes, through the source/destination images */
IppiSize roiSize = { WIDTH, HEIGHT }; /* Size of source/destination ROI in pixels */
Ipp32u kernelSize = 3;
Ipp32f sigma = 0.56f;
Ipp8u *pBuffer = NULL; /* Pointer to the work buffer */
IppFilterGaussianSpec* pSpec = NULL; /* context structure */
int iTmpBufSize = 0, iSpecSize = 0; /* Common work buffer size */
IppiBorderType borderType = ippBorderRepl;
Ipp8u borderValue = 0;
int numChannels = 1;

psrc=ippiMalloc_8u_C1(roiSize.width, roiSize.height, &srcStep);
pDst = ippiMalloc_8u_C1(roiSize.width, roiSize.height, &dstStep);

/* fill source image with const value = 5*/
for(int i=0;i<HEIGHT; i++)
for(int j=0; j<WIDTH; j++)
*(pSrc + (i * srcStep/sizeof(char)) + j ) = 5;

status = ippiFilterGaussianGetBufferSize(roiSize, kernelSize, ipp8u,
numChannels, &iSpecSize, &iTmpBufSize);
if(status != ippStsNoErr)
return 0;

pSpec = (IppFilterGaussianSpec *)ippsMalloc_8u(iSpecSize);
pBuffer = ippsMalloc_8u(iTmpBufSize);

status = ippiFilterGaussianInit(roiSize, kernelSize, sigma,
borderType, ipp8u, numChannels, pSpec, pBuffer);
if(status != ippStsNoErr)
return 0;

status = ippiFilterGaussianBorder_8u_C1R(pSrc, srcStep, pDst, dstStep,
roiSize, borderValue, pSpec, pBuffer);
if(status != ippStsNoErr)
return 0;

// Print output
cout<<"\nsrc"<<endl;
cout<<"------"<<endl;
for(int i=0;i<HEIGHT; i++)
{
for(int j=0; j<WIDTH; j++)
{
cout<<" "<<int(*(pSrc + (i * srcStep/sizeof(char)) + j ));
}
cout<<endl;
}
cout<<"\ndst"<<endl;
cout<<"------"<<endl;
for(int i=0;i<HEIGHT; i++)
{
for(int j=0; j<WIDTH; j++)
{
cout<<" "<<int(*(pDst + (i * srcStep/sizeof(char)) + j ));
}
cout<<endl;
}

ippsFree(pBuffer);
ippsFree(pSpec);
ippiFree(pSrc);
ippiFree(pDst);

return (int)status;
}

[VidyalathaB_Intel]

Hi,

Thanks for reaching out to us.

The issue is reproducible from our end.

We are looking into this issue internally. we will get back to you soon.

Regards,

Vidya.

[Gennady_F_Intel]

Eyal,

the fix of the problem could be available into the latest version of IPP ( 2021.5).

Here is the output I see while linked your case against this version of IPP:

major = 2021

minor = 5

majorBuild = 0

build = -1274761199

targetCpu = l9  

Name = ippSP AVX2 (l9)

Version = 2021.5 (r0xb404b011)

BuildDate = Nov 18 2021

src

------

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

 5 5 5 5 5 5 5 5 5 5

dst

------

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

 5 5 5 5 5 5 5 5 4 4

-Gennady