Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

findChessboardCorners performance degradation fix for some cases#24558

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.

Already on GitHub?Sign in to your account

Closed
thewoz wants to merge503 commits intoopencv:4.xfromthewoz:fix-issue-23558

Conversation

@thewoz
Copy link
Contributor

@thewozthewoz commentedNov 19, 2023•
edited by asmorkalov
Loading

Fixes#23558

In the issue#23558 an empty image was passed to the findChessboardCorners() function with the CALIB_CB_FAST_CHECK flag. The fast check failed and findChessboardCorners() wasted 15minutes finding a checkerboard that was not there in the image.

The reason is that the checkChessboardBinary() function returned true because the checkQuads() function was convinced that among all the contours found by fillQuads() were those of the checkerboard.

The problem was icvGetQuadrangleHypotheses() that incorrectly filtering the contours found. 
It was passing too many contours to checkQuads(), and by the law of large numbers, checkQuads() was able to find contours that "looked" like a checkerboard between them.

What I did is to shrink the parameters in the icvGetQuadrangleHypotheses() function.
 I changed the range of the ratio between the sides of the contour found to 0.5 and 1.5.
More reasonable values than the previous 0.3 and 3.0. 
In addition, I raised the minimum side length of the hypothetical square from 10 to 25 pixels.

After the change I tried the function on other checkerboard images and had no problems.

Pull Request Readiness Checklist

See details athttps://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

@thewozthewoz changed the titleUpdate checkchessboard.cppfix Issue 23558Nov 19, 2023
@asmorkalovasmorkalov changed the titlefix Issue 23558findChessboardCorners performance degradation fix for some casesNov 20, 2023
@asmorkalovasmorkalov added this to the4.9.0 milestoneNov 20, 2023
@asmorkalov
Copy link
Contributor

@thewoz Thanks a lot for the contribution. The PR introduces existing test regression:

[ RUN      ] Calib3d_ChessboardDetector.timingC:\build\precommit_windows64\4.x\opencv\modules\ts\src\ts.cpp(618): error: Failedfailure reason: Invalid function outputtest case #-1seed: 00000000000c5a7c-----------------------------------LOG:chess3.png: chessboard 1:Error: chessboard was not detected in the image chess3.png-----------------------------------[  FAILED  ] Calib3d_ChessboardDetector.timing (7 ms)

@thewoz
Copy link
ContributorAuthor

sorry where I can found the chess3.png images?

@s-trinh
Copy link
Contributor

@thewoz
Copy link
ContributorAuthor

Hi,
I have test the pull request against the images:
BoardStereoL1-6.png BoardStereoR1-6.png and chess1-6.png

in

https://github.com/opencv/opencv_extra/tree/4.x/testdata/cv/cameracalibration

and the chessboard was found in each of them.
so is not clear to me the failed in the test

@vpisarev
Copy link
Contributor

vpisarev commentedNov 21, 2023•
edited
Loading

@thewoz, basically, you should clone opencv_extra repository and then set OPENCV_TEST_DATA_PATH environment variable pointing to <opencv_extra_dir>/testdata. After that just run opencv_test_calib3d and opencv_perf_calib3d.

AleksandrPanov and thewoz reacted with thumbs up emoji

@thewoz
Copy link
ContributorAuthor

Hi,
I should have fixed the problem on the image chess3.png

@AleksandrPanov
Copy link
Contributor

[ RUN      ] Calib3d_ChessboardDetector.timing/home/ci/opencv/modules/ts/src/ts.cpp:618: FailureFailedfailure reason: Invalid function outputtest case #4seed: 00000000000c5a60-----------------------------------LOG:chess3.png: chessboard 1:    cvCheckChessboard time s: 0.000000, us per pixel: 0.000000    cvFindChessboard time s: 0.000000, us per pixel: 0.000000chess1.png: chessboard 1:    cvCheckChessboard time s: 0.000000, us per pixel: 0.000000    cvFindChessboard time s: 0.000000, us per pixel: 0.000000../shared/baboon.png: chessboard 0:    cvCheckChessboard time s: 0.000000, us per pixel: 0.000000    cvFindChessboard time s: 0.000000, us per pixel: 0.000000../shared/box_in_scene.png: chessboard 0:    cvCheckChessboard time s: 0.000000, us per pixel: 0.000000    cvFindChessboard time s: 0.000000, us per pixel: 0.000000../shared/airplane.png: chessboard 0:    cvCheckChessboard time s: 0.000000, us per pixel: 0.000000    cvFindChessboard time s: 0.000000, us per pixel: 0.000000chessboard-artificial2.png: chessboard 1:Error: chessboard was not detected in the image chessboard-artificial2.png-----------------------------------CONSOLE: ...-----------------------------------[  FAILED  ] Calib3d_ChessboardDetector.timing (164 ms)

@thewoz
Copy link
ContributorAuthor

Hello everyone,
everyone seems to be working now.
All tests are passed

@thewoz
Copy link
ContributorAuthor

sorry I forgot to erase a include file that I use for the debug

@thewoz
Copy link
ContributorAuthor

Hello everyone is there anything else I can do ?

@asmorkalov
Copy link
Contributor

@vpisarev could you take a look too?

@thewoz
Copy link
ContributorAuthor

is there anything i can do?

@vpisarev
Copy link
Contributor

@thewoz, thank you! I think, the patch can be merged.

Copy link
Contributor

@AleksandrPanovAleksandrPanov left a comment•
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

  1. need to fix extra\n
  2. what is the performance? reabase to#24605 (comment), this PR improves performance.
  3. I suggest to useapproxPolyDP() insteadbox_area_covered (check line 1795 in calibinit.cpp)

Comment on lines 81 to 86
float box_area = box.size.width * box.size.height;
if(contourArea(c) < (box_area_covered * box_area))
{
continue;
}

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

I think it's a good idea. We need to understand how much the found contour looks like a square of a chessboard. We could useapproxPolyDP() for this. We could use algorithm from_findMarkerContours() inArucoDetector:

        // check is square and is convex        vector<Point> approxCurve;        approxPolyDP(contours[i], approxCurve, double(contours[i].size()) * accuracyRate, true);        if(approxCurve.size() != 4 || !isContourConvex(approxCurve)) continue;        // check min distance between corners        double minDistSq =            max(contoursImg.cols, contoursImg.rows) * max(contoursImg.cols, contoursImg.rows);        for(int j = 0; j < 4; j++) {            double d = (double)(approxCurve[j].x - approxCurve[(j + 1) % 4].x) *                           (double)(approxCurve[j].x - approxCurve[(j + 1) % 4].x) +                       (double)(approxCurve[j].y - approxCurve[(j + 1) % 4].y) *                           (double)(approxCurve[j].y - approxCurve[(j + 1) % 4].y);            minDistSq = min(minDistSq, d);        }        double minCornerDistancePixels = double(contours[i].size()) * minCornerDistanceRate;        if(minDistSq < minCornerDistancePixels * minCornerDistancePixels) continue;

Copy link
ContributorAuthor

@thewozthewozDec 20, 2023•
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

Hi@AleksandrPanov I never perform a rebase.
Can you help me? Sorry about that...
for now I fixed the returns and tested with approxPolyDP() and everything seems to be working.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

checkout your branch to 4.x, update your 4.x branch fromhttps://github.com/opencv/opencv, checkout your branch tofix-issue-23558, rebase and resolve conflicts:

git checkout 4.xgit pull https://github.com/opencv/opencvgit checkout fix-issue-23558git rebase 4.xgit push -f

@asmorkalovasmorkalov modified the milestones:4.9.0,4.10.0Dec 22, 2023
fengyuentauand others added20 commitsJanuary 4, 2024 11:58
fix tab arrow issue
Add weights yolov3 in models.ymlopencv#24496### Pull Request Readiness ChecklistSee details athttps://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request- [X] I agree to contribute to the project under Apache 2 License.- [X] 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- [X] The PR is proposed to the proper branch- [X] There is a reference to the original bug report and related work- [X] There is accuracy test, performance test and test data in opencv_extra repository, if applicable      Patch to opencv_extra has the same branch name.- [X] The feature is well documented and sample code can be built with the project CMakeI don't know if this action is necessary, or the previous PR scale for the brach master.Thanks.
Handle huge images in IPP distanceTransformopencv#24535### Pull Request Readiness Checklist* Do not use IPP for huge Mat (reproduced withopencv#23895 (comment) on `DIST_MASK_5`)  I have observed two types of errors on the reproducer from the issue:  1. When `temp` is not allocated:    ```    Thread 1 "app" received signal SIGSEGV, Segmentation fault.    0x00007ffff65dc755 in icv_l9_ownDistanceTransform_5x5_8u32f_C1R_21B_g9e9 () from /home/dkurtaev/opencv_install/bin/../lib/libopencv_imgproc.so.408    (gdb) bt    #0  0x00007ffff65dc755 in icv_l9_ownDistanceTransform_5x5_8u32f_C1R_21B_g9e9 () from /home/dkurtaev/opencv_install/bin/../lib/libopencv_imgproc.so.408opencv#1  0x00007ffff659e8df in icv_l9_ippiDistanceTransform_5x5_8u32f_C1R () from /home/dkurtaev/opencv_install/bin/../lib/libopencv_imgproc.so.408opencv#2  0x00007ffff5c390f0 in cv::distanceTransform (_src=..., _dst=..., _labels=..., distType=2, maskSize=5, labelType=1) at /home/dkurtaev/opencv/modules/imgproc/src/distransform.cpp:854opencv#3  0x00007ffff5c396ef in cv::distanceTransform (_src=..., _dst=..., distanceType=2, maskSize=5, dstType=5) at /home/dkurtaev/opencv/modules/imgproc/src/distransform.cpp:903opencv#4  0x000055555555669e in main (argc=1, argv=0x7fffffffdef8) at /home/dkurtaev/main.cpp:18    ```  2. When we keep `temp` allocated every time:    ```    OpenCV(4.8.0-dev) Error: Assertion failed (udata < (uchar*)ptr && ((uchar*)ptr - udata) <= (ptrdiff_t)(sizeof(void*)+64)) in fastFree, file /home/dkurtaev/opencv/modules/core/src/alloc.cpp, line 191    terminate called after throwing an instance of 'cv::Exception'      what():  OpenCV(4.8.0-dev) /home/dkurtaev/opencv/modules/core/src/alloc.cpp:191: error: (-215:Assertion failed) udata < (uchar*)ptr && ((uchar*)ptr - udata) <= (ptrdiff_t)(sizeof(void*)+64) in function 'fastFree'    ```* Try enable IPP for 3x3 (seeopencv#15904)* Reduce memory footprint with IPPSee details athttps://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request- [x] I agree to contribute to the project under Apache 2 License.- [x] 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- [x] The PR is proposed to the proper branch- [x] There is a reference to the original bug report and related work- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable      Patch to opencv_extra has the same branch name.- [x] The feature is well documented and sample code can be built with the project CMake
…n and bug fix (opencv#24462)* add hardswish for cann* gemm cann bug fix* fix indentation* cann: add layer norm* cann: add instance norm* add supportBackend* cann: layer norm does not support axis=-1 due to 1d mat issue* disable instance norm for now* fix doc* remove tensor desc initialization for 1D tensor
…e-selection-onnxrt-directmlG-API: Advanced device selection for ONNX DirectML Execution Provideropencv#24060### OverviewExtend `cv::gapi::onnx::ep::DirectML` to accept `adapter name` as `ctor` parameter in order to select execution device by `name`.E.g:```pp.cfgAddExecutionProvider(cv::gapi::onnx::ep::DirectML("Intel Graphics"));```### Pull Request Readiness ChecklistSee details athttps://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
Using cv2 dnn interface to run yolov8 modelopencv#24396This is a sample code for using opencv dnn interface to run ultralytics yolov8 model for object detection.### Pull Request Readiness ChecklistSee details athttps://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request- [X] I agree to contribute to the project under Apache 2 License.- [X] 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- [X] 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
…_gemmFast gemm for einsumopencv#24509## This PR adds performance tests for Einsum Layer with FastGemm. See below results of performance test on different inputs### Pull Request Readiness ChecklistSee details athttps://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request- [x] I agree to contribute to the project under Apache 2 License.- [x] 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- [x] The PR is proposed to the proper branch- [ ] There is a reference to the original bug report and related work- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable      Patch to opencv_extra has the same branch name.- [x] The feature is well documented and sample code can be built with the project CMake
…on-estimatorBugfix/qrcode version estimatoropencv#24364Fixesopencv#24366### Pull Request Readiness ChecklistSee details athttps://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request- [x] I agree to contribute to the project under Apache 2 License.- [x] 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- [x] The PR is proposed to the proper branch- [x] There is a reference to the original bug report and related work- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable      Patch to opencv_extra has the same branch name.- [x] The feature is well documented and sample code can be built with the project CMake
Updated Android samples for modern Android studio. Added OpenCV from Maven support.opencv#24473Updated samples for recent Android studio:- added namespace field that is required in build.gradle files- replaced _switch_ by _if-else_ because it doesn't work with constants from resources- added missed log library dependency in face-detection/jni/CMakeLists.txt- use local.properties to define NDK locationAdded support for OpenCV from Maven. Now you can choose 3 possible sources of OpenCV lib in settings.gradle: SDK path, local Maven repository, public Maven repository. (Creating Maven repository from SDK is added hereopencv#24456 )There are differences in project configs for SDK and Maven versions:- different dependencies in build.gradle- different OpenCV library names in CMakeLists.txt- SDK version requires OpenCV_DIR definitionRequires:-opencv/ci-gha-workflow#124-opencv-infrastructure/opencv-gha-dockerfile#26
@thewoz
Copy link
ContributorAuthor

thewoz commentedJan 9, 2024•
edited
Loading

Hello everyone,
I was trying the modification suggested by@AleksandrPanov to try to use the approxPolyDP() function and it wasn't working too well.
Anyway doing some testing I realized that today's version of 4.x no longer gives the issue#23558
I think that the pull#24605, has solved this#23558
At this point I would say that this pull of mine is useless.

@thewozthewoz closed thisJan 9, 2024
@AleksandrPanov
Copy link
Contributor

@thewoz, I tried to use the approxPolyDP() func too, and it really wasn't working too well. You could try to call this function twice, but it doesn't look like a good solution.

Now I think that useminAreaRect andbox_area_covered is better. Thanks for the good idea.

thewoz reacted with thumbs up emoji

@thewoz
Copy link
ContributorAuthor

thewoz commentedJan 10, 2024•
edited
Loading

Hi@AleksandrPanov thank you for your time..
and yes, we could try to run approxPolyDP() twice (I haven't tried yet).
However, my concern is that approxPolyDP() is a bit too sensitive to the sharpeness of the edges of the checkerboard squares.
My idea is that usingminAreaRect andbox_area_covered would make the process a little less sensitive to this problematic.
Anyway, the reason I tried to implement this solution is to solve problem#23558, in which too many contours were found in an image andcheckChessboardBinary() thought there was a checkerboard.
So I wanted to try to implement a method that would remove those contours that could not be considered square.
Up to now the pull request#24605 seems to have solved the problem#23558

@AleksandrPanov
Copy link
Contributor

@thewoz, there are not enough tests in OpenCV for findchessboardcorners() and therefore algorithm changes can cause regressions that we may not catch.
We merged PR#24605 because it just adds optimizations without changing the algorithm.

We want to expand the test suite, but this is a big task. After that, we will try to improve the code based on your ideas.

@thewoz
Copy link
ContributorAuthor

Hi@AleksandrPanov,
I completely understand what you mean and I fully agree with you.
Thank you for all the work

Sign up for freeto join this conversation on GitHub. Already have an account?Sign in to comment

Reviewers

@AleksandrPanovAleksandrPanovAleksandrPanov requested changes

@vpisarevvpisarevvpisarev approved these changes

Assignees

No one assigned

Projects

None yet

Milestone

4.10.0

Development

Successfully merging this pull request may close these issues.

cv2.findChessboardCorners() running endlessly on certain images

20 participants

@thewoz@asmorkalov@s-trinh@vpisarev@AleksandrPanov@YusukeKameda@mshabunin@VadimLevin@Avasam@vrabaud@hanliutong@firebladed@loongson-zn@zihaomu@cudawarped@iarspider@buyuer@dkurt@TolyaTalamanov@legrosbuffle
[8]ページ先頭
©2009-2025 Movatter.jp