Merge pull request #22959 from feuerste:parallel_mertens

Parallelize implementation of HDR MergeMertens.

* Parallelize MergeMertens.

* Added performance tests for HDR.

* Ran clang-format.

* Optimizations.

* Fix data path for Windows.

* Remove compiiation warning on Windows.

* Remove clang-format for existing file.

* Addressing reviewer comments.

* Ensure correct summation order.

* Add test for determinism.

* Move result pyramid into sync struct.

* Reuse sync for first loop as well.

* Use OpenCV's threading primitives.

* Remove cout.

modules/python/test/test_umat.py

@@ -107,12 +107,19 @@ class UMat(NewOpenCVTests):
 
         images, _ = load_exposure_seq(os.path.join(test_data_path, 'exposures'))
 
+        # As we want to test mat vs. umat here, we temporarily set only one worker-thread to achieve
+        # deterministic summations inside mertens' parallelized process.
+        num_threads = cv.getNumThreads()
+        cv.setNumThreads(1)
+
         merge = cv.createMergeMertens()
         mat_result = merge.process(images)
 
         umat_images = [cv.UMat(img) for img in images]
         umat_result = merge.process(umat_images)
 
+        cv.setNumThreads(num_threads)
+
         self.assertTrue(np.allclose(umat_result.get(), mat_result))