Skip to content

Add warpAffine IPPIW implementation to replace with old version#22074

Merged
asmorkalov merged 1 commit intoopencv:4.xfrom
bwang30:opencv-warpAffine-ippiw
Oct 4, 2022
Merged

Add warpAffine IPPIW implementation to replace with old version#22074
asmorkalov merged 1 commit intoopencv:4.xfrom
bwang30:opencv-warpAffine-ippiw

Conversation

@bwang30
Copy link
Copy Markdown
Contributor

@bwang30 bwang30 commented Jun 7, 2022

The old version of IPP warpaffine can't work(can't pass build) which was controlled by flag IPP_DISABLE_WARPAFFINE.
This pull request provides IPPIW warpaffine version controlled by flag HAVE_IPP_IW , and check ipp NE flag for the different result issue between OpenCV and IPP.

Tested 100 images with src size is 1920x1080, data type is CV_8UC3, ROI is the whole image:

image

code snippet as following:

const int bordertype = cv::BORDER_CONSTANT;
int interpolation = cv::INTER_LINEAR; 
double angle = 30;                    

vector<cv::Mat> srcData;
for (auto imgpath : sInputFiles)
{
    Mat img = cv::imread(imgpath, cv::IMREAD_COLOR);
    cv::Size dsize(1920, 1080);
    Mat img_resized;
    cv::resize(img, img_resized, dsize, 0, 0, cv::INTER_CUBIC);
    srcData.push_back(img);
  }

vector<cv::Mat> dstData(srcData.size());
...
cv::Mat M = cv::getRotationMatrix2D(center, rotateAngle, scale);

double radianAngle = rotateAngle * 3.14159 / 180;
int drows = int(cols * fabs(sin(radianAngle)) + rows * fabs(cos(radianAngle)));
int dcols = int(rows * fabs(sin(radianAngle)) + cols * fabs(cos(radianAngle)));
M.at<double>(0, 2) += (dcols - cols) / 2;
M.at<double>(1, 2) += (drows - rows) / 2;

cv::warpAffine(src, dst, M, Size(dcols, drows), interpolation, bordertype, Scalar(0, 0, 0));

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

  • I agree to contribute to the project under Apache 2 License.
  • To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
  • The PR is proposed to the proper branch
  • There is a reference to the original bug report and related work
  • There is accuracy test, performance test and test data in opencv_extra repository, if applicable
    Patch to opencv_extra has the same branch name.
  • The feature is well documented and sample code can be built with the project CMake

@bwang30 bwang30 changed the title Add warpAffine IPPIW implementation protected by ipp NE flag Add warpAffine IPPIW implementation to replace with old version Jun 7, 2022
@asmorkalov asmorkalov requested a review from vpisarev October 3, 2022 13:00
@vpisarev
Copy link
Copy Markdown
Contributor

vpisarev commented Oct 3, 2022

@bwang30, thank you! 👍

@bwang30 @asmorkalov
Add warpAffine IPPIW implementation protected by ipp NE flag
ed3b56d 
Signed-off-by: robin <bin.wang@intel.com>
@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Oct 4, 2022

Rebased to 4.x and squashed commits.

@asmorkalov asmorkalov force-pushed the opencv-warpAffine-ippiw branch from ba5dd0f to ed3b56d Compare October 4, 2022 05:42
@asmorkalov asmorkalov added this to the 4.7.0 milestone Oct 4, 2022
@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Oct 4, 2022 •
edited
Loading

OpenCV performance test status:

Geometric mean (ms)

                              Name of Test                               warpaffine warpaffine warpaffine
                                                                          default      ipp        ipp    
                                                                                                   vs    
                                                                                               warpaffine
                                                                                                default  
                                                                                               (x-factor)
WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT)       0.573      0.518       1.11   
WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE)      0.560      0.429       1.31   
WarpAffine::TestWarpAffine::(640x480, INTER_NEAREST, BORDER_CONSTANT)      0.411      0.303       1.36   
WarpAffine::TestWarpAffine::(640x480, INTER_NEAREST, BORDER_REPLICATE)     0.247      0.307       0.80   
WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT)      0.582      0.579       1.01   
WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE)     1.128      1.117       1.01   
WarpAffine::TestWarpAffine::(1280x720, INTER_NEAREST, BORDER_CONSTANT)     0.515      0.411       1.25   
WarpAffine::TestWarpAffine::(1280x720, INTER_NEAREST, BORDER_REPLICATE)    0.465      0.496       0.94   
WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT)     1.263      1.141       1.11   
WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE)    3.937      3.717       1.06   
WarpAffine::TestWarpAffine::(1920x1080, INTER_NEAREST, BORDER_CONSTANT)    0.888      0.861       1.03   
WarpAffine::TestWarpAffine::(1920x1080, INTER_NEAREST, BORDER_REPLICATE)   1.016      1.001       1.02

System: AMD Ryzen 7 2700X (8 cores), 64GB RAM.

@alalek
Copy link
Copy Markdown
Member

alalek commented Oct 4, 2022

Results above show performance measurement instability.
Optimization code is guarded by useIPP_NotExact() flag.
Extra steps are required, like OPENCV_IPP=NE env should be used.

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Oct 4, 2022

Yes, I've already found it.

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Oct 4, 2022

New statistics with IPP NE forced to true:

WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT)       0.573      0.256       2.24   
WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE)      0.560      0.226       2.48   
WarpAffine::TestWarpAffine::(640x480, INTER_NEAREST, BORDER_CONSTANT)      0.411      0.114       3.62   
WarpAffine::TestWarpAffine::(640x480, INTER_NEAREST, BORDER_REPLICATE)     0.247      0.058       4.23   
WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT)      0.582      failed       -     
WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE)     1.128      0.771       1.46   
WarpAffine::TestWarpAffine::(1280x720, INTER_NEAREST, BORDER_CONSTANT)     0.515      failed       -     
WarpAffine::TestWarpAffine::(1280x720, INTER_NEAREST, BORDER_REPLICATE)    0.465      0.159       2.93   
WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT)     1.263      failed       -     
WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE)    3.937      1.307       3.01   
WarpAffine::TestWarpAffine::(1920x1080, INTER_NEAREST, BORDER_CONSTANT)    0.888      failed       -     
WarpAffine::TestWarpAffine::(1920x1080, INTER_NEAREST, BORDER_REPLICATE)   1.016      0.311       3.27

@asmorkalov asmorkalov merged commit 8f0edf6 into opencv:4.x Oct 4, 2022
@xethan666
Copy link
Copy Markdown

xethan666 commented Oct 9, 2022

New statistics with IPP NE forced to true:

WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT)       0.573      0.256       2.24   
WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE)      0.560      0.226       2.48   
WarpAffine::TestWarpAffine::(640x480, INTER_NEAREST, BORDER_CONSTANT)      0.411      0.114       3.62   
WarpAffine::TestWarpAffine::(640x480, INTER_NEAREST, BORDER_REPLICATE)     0.247      0.058       4.23   
WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT)      0.582      failed       -     
WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE)     1.128      0.771       1.46   
WarpAffine::TestWarpAffine::(1280x720, INTER_NEAREST, BORDER_CONSTANT)     0.515      failed       -     
WarpAffine::TestWarpAffine::(1280x720, INTER_NEAREST, BORDER_REPLICATE)    0.465      0.159       2.93   
WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT)     1.263      failed       -     
WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE)    3.937      1.307       3.01   
WarpAffine::TestWarpAffine::(1920x1080, INTER_NEAREST, BORDER_CONSTANT)    0.888      failed       -     
WarpAffine::TestWarpAffine::(1920x1080, INTER_NEAREST, BORDER_REPLICATE)   1.016      0.311       3.27

which HW platform this result based on, also AMD Ryzen 7 2700X (8 cores), 64GB RAM?

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Oct 10, 2022

Yes.

@alalek alalek mentioned this pull request Jan 8, 2023
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@vpisarev vpisarev vpisarev approved these changes

Assignees

@vpisarev vpisarev

Labels

category: imgproc optimization

Projects

None yet

Milestone

4.7.0

Development

Successfully merging this pull request may close these issues.

5 participants

@bwang30 @vpisarev @asmorkalov @alalek @xethan666