Transform offset to indeces for MatND in minMaxIdx HAL #25563
Conversation
|
@mplatings please take a look. |
Co-authored-by: Michael Platings <[email protected]>
Transform offset to indeces for MatND in minMaxIdx HAL opencv#25563 Address comments in opencv#25553 ### Pull Request Readiness Checklist See details at https://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 - [ ] 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
| if (res == CV_HAL_ERROR_OK) | ||
| { | ||
| if (minIdx) | ||
| ofs2idx(src, minIdx[0], minIdx); |
There was a problem hiding this comment.
@asmorkalov I've done some more testing and I believe it should be minIdx[1] and maxIdx[1]. I thought that minIdx[0] would correspond to the X direction, but it seems that in OpenCV it corresponds to the row. According do the docs:
single-row matrix is 1xN matrix (and therefore minIdx/maxIdx will be (0,j1)/(0,j2))
If I change these indices from 0 to 1 then my tests pass.
There was a problem hiding this comment.
@mplatings It's not clear from the docs if minIdx and maxIdx should contain pointers to an offset to convert it to array indices or to indices themselves.
In the first case (which is how OpenCV tab functions from getMinmaxTab() actually work) there should be 0th index and no such thing as minIdx[i] or maxIdx[i] is ever possible for every i > 0.
In the second case you're right since HAL call gets 1-dimensional array and puts Y index to minIdx[0] and X to minIdx[1] (the same for maxIdx). Since it's just one dimensional, Y index should always be zero which looks impractical to me but looks like this is how it is implemented now.
I think we should choose between these two options first.
By the way, can you tell me what tests are failing? Looks like OpenCV tests do not cover that case.
There was a problem hiding this comment.
KleidiCV's tests fail with the head of the OpenCV 4.x branch. Specifically this: https://gitlab.arm.com/kleidi/kleidicv/-/blob/main/conformity/opencv/test_min_max.cpp?ref_type=heads#L59
In KleidiCV 0.1.0 we did not enable minmax in the HAL so from our point of view it's fine for us to change how the HAL should work. If the HAL code should return an offset instead of 2-dimensional indices then that makes KleidiCV's code simpler.
So I'm happy to either change these indices from 0 to 1, or to require that the HAL returns an offset and we will change KleidiCV to match.
There was a problem hiding this comment.
OK, let's do like this so that HAL should return offset.
I'll prepare PR to OpenCV which documents such behaviour.
Transform offset to indeces for MatND in minMaxIdx HAL opencv#25563 Address comments in opencv#25553 ### Pull Request Readiness Checklist See details at https://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 - [ ] 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
Transform offset to indeces for MatND in minMaxIdx HAL opencv#25563 Address comments in opencv#25553 ### Pull Request Readiness Checklist See details at https://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 - [ ] 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
Transform offset to indeces for MatND in minMaxIdx HAL opencv#25563 Address comments in opencv#25553 ### Pull Request Readiness Checklist See details at https://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 - [ ] 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
Address comments in #25553
Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
Patch to opencv_extra has the same branch name.