I found that in some cases, Otsu's method can have two or more equal thresholds. In case of the CUDA implementation, this would create a race-condition where the last thread to write the output gets to choose which Otsu threshold to use. The resulting image from cv::cuda::threshold would be the same, but the returned threshold would vary for each run.

I solve this by first doing a poll to check if there are multiple matches with __syncthreads_count. If not, we can return the threshold as before. If there are more than one, the kernel does a minimum reduction to find the lowest threshold that matches the otsu score. It doesn't matter which threshold you choose as long as it is consistent, so I choose the smallest one.

I'm unsure when __syncthreads_count as introduced, but it is compatible with CC≥2.0. CUDA Toolkit's archive only goes back to version 8.0, but it was documented back then (https://docs.nvidia.com/cuda/archive)

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