Merge pull request #26124 from fengyuentau:dnn/topk_dtype

dnn(5.x): handle topk data type #26124

Resolves https://github.com/opencv/opencv/issues/26076

### 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
- [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

modules/dnn/test/test_onnx_conformance_layer_parser_denylist.inl.hpp

@@ -395,7 +395,7 @@
 "test_tfidfvectorizer_tf_uniandbigrams_skip5", // Issue:: Parser: Can't create layer "onnx_node_output_0!Y" of type "TfIdfVectorizer" in function 'getLayerInstance'
 "test_tile", // Issue:: Parser: ONNX/Tile: repeats being non-constant is not supported. in function 'parseTile' (layer parameters are dynamic)
 "test_tile_precomputed", //  // ---- same as above ---
-"test_top_k", // Issue:: Parser: Can't create layer "onnx_node_output_0!values" of type "TopK" in function 'getLayerInstance'
+"test_top_k", // Issue:: K being input is not compatible with the current engine
 "test_top_k_negative_axis",  // ---- same as above ---
 "test_top_k_smallest",  // ---- same as above ---
 "test_training_dropout", // Issue::cvtest::norm::wrong data type

modules/dnn/test/test_onnx_importer.cpp

@@ -3278,8 +3278,12 @@ TEST_P(Test_ONNX_layers, ClipDivSharedConstant) {
     testONNXModels("clip_div_shared_constant");
 }
 
-// Bug: https://github.com/opencv/opencv/issues/26076
-TEST_P(Test_ONNX_layers, DISABLED_TopK) {
+TEST_P(Test_ONNX_layers, TopK) {
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH ||
+        backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
+        backend == DNN_BACKEND_INFERENCE_ENGINE) {
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE); // OpenVINO does not support int64
+    }
     auto test = [&](const std::string &basename, double l1 = 0, double lInf = 0) {
         std::string onnxmodel = _tf("models/" + basename + ".onnx", true);
         Mat input = readTensorFromONNX(_tf("data/input_" + basename + ".pb"));
@@ -3299,8 +3303,6 @@ TEST_P(Test_ONNX_layers, DISABLED_TopK) {
         Mat output_res_val = outputs.front(),
             output_res_ind = outputs.back();
 
-        output_ref_ind.convertTo(output_ref_ind, CV_32F); // TODO: revise this conversion in 5.x
-
         normAssert(output_ref_val, output_res_val, (basename + " values").c_str(), l1 ? l1 : default_l1, lInf ? lInf : default_lInf);
         normAssert(output_ref_ind, output_res_ind, (basename + " indices").c_str(), l1 ? l1 : default_l1, lInf ? lInf : default_lInf);