Merge pull request #23319 from fengyuentau:fix_zoo_issue_136

Related issue: https://github.com/opencv/opencv_zoo/issues/136

Features added:

- Support operators with multiple output: ONNX Split.
- Support Slice without steps.

Bugs fixed:

- Wrong settings in ClipByValue (Relu6).
- Wrong calculation of pads in convolution layer (It is wrong generally but only fixed specifically for CANN for now).

### 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/src/net_cann.cpp

@@ -124,6 +124,30 @@ void NetImplCann::initBackend(const std::vector<LayerPin>& blobsToKeep_)
     if (!newWasSupported)
         return ;
 
+    // initialize each blob wrappers' names
+    for (MapIdToLayerData::const_iterator it = layers.begin(); it != layers.end(); ++it)
+    {
+        const LayerData& ld = it->second;
+        if (ld.id == 0)
+        {
+            for (int i = 0; i < ld.outputBlobsWrappers.size(); ++i)
+            {
+                auto cannWrapper = ld.outputBlobsWrappers[i].dynamicCast<CannBackendWrapper>();
+                // cannWrapper->name = netInputLayer->outNames.empty() ? cv::format("%s_%d", ld.name.c_str(), i) : netInputLayer->outNames[i];
+                cannWrapper->name = std::string("y");
+            }
+        }
+        else
+        {
+            for (int i = 0; i < ld.outputBlobsWrappers.size(); ++i)
+            {
+                auto cannWrapper = ld.outputBlobsWrappers[i].dynamicCast<CannBackendWrapper>();
+                // cannWrapper->name = ld.outputBlobsWrappers.size() > 1 ? (ld.name + ":" + std::to_string(i)) : ld.name;
+                cannWrapper->name = ld.outputBlobsWrappers.size() > 1 ? (std::string("y") + std::to_string(i)) : std::string("y");
+            }
+        }
+    }
+
     // convert layers to CANN operators,
     // collect graph input and output operators,
     // collect and input and output wrappers
@@ -141,15 +165,16 @@ void NetImplCann::initBackend(const std::vector<LayerPin>& blobsToKeep_)
         {
             for (int i = 0; i < ld.outputBlobsWrappers.size(); i++)
             {
-                std::string inputName = netInputLayer->outNames.empty() ? cv::format("%s_%d", ld.name.c_str(), i) : netInputLayer->outNames[i];
-                auto inputOp = std::make_shared<ge::op::Data>(inputName);
-
                 // retrieve tensor description
                 auto wrapper = ld.outputBlobsWrappers[i];
                 graphInputWrappers.push_back(wrapper);
                 auto cannWrapper = wrapper.dynamicCast<CannBackendWrapper>();
                 CV_Assert(!cannWrapper.empty());
 
+                // create graph input op
+                std::string inputOpName = netInputLayer->outNames.empty() ? cv::format("%s_%d", ld.name.c_str(), i) : netInputLayer->outNames[i];
+                auto inputOp = std::make_shared<ge::op::Data>(inputOpName);
+
                 inputOp->update_input_desc_x(*(cannWrapper->desc_));
                 inputOp->update_output_desc_y(*(cannWrapper->desc_));
 
@@ -170,14 +195,14 @@ void NetImplCann::initBackend(const std::vector<LayerPin>& blobsToKeep_)
                 {
                     layerInputNodes.push_back(netInputNodes[layerInputOid]);
                 }
-                else // here we do not consider an op with multiple outputs
+                else
                 {
                     layerInputNodes.push_back(layers[layerInputLid].backendNodes[preferableBackend]);
                 }
             }
 
             CV_LOG_INFO(NULL, "DNN/CANN: converting layer " << ld.name << "@" << ld.type << "@" << ld.id << " to CANN operator");
-            auto backendNode = layer->initCann(ld.inputBlobsWrappers, ld.id, layerInputNodes);
+            auto backendNode = layer->initCann(ld.inputBlobsWrappers, layerInputNodes); // it's ok if ld.name is empty
 
             // collect outputs
             bool isOutputNode = ld.consumers.size() == 0 ? true : false;
@@ -201,7 +226,7 @@ void NetImplCann::initBackend(const std::vector<LayerPin>& blobsToKeep_)
 
     // build graph from collected graph inputs and outputs
     CV_LOG_INFO(NULL, "DNN/CANN: building ge::Graph");
-    std::string graphName = cv::format("graph_%d", 0);
+    std::string graphName = cv::format("graph_%d", networkId);
     std::shared_ptr<ge::Graph> graph = std::make_shared<ge::Graph>(graphName.c_str());
     (void)graph->SetInputs(graphInputOps);
     (void)graph->SetOutputs(graphOutputOps);
@@ -292,9 +317,9 @@ std::shared_ptr<ge::ModelBufferData> compileCannGraph(std::shared_ptr<ge::Graph>
 
 #if 0
             // (optional). Dump model
-            AscendString graph_name;
-            graph.GetName(graph_name);
-            aclgrphDumpGraph(graph, graph_name.GetString(), 7);
+            ge::AscendString graph_name;
+            graph->GetName(graph_name);
+            aclgrphDumpGraph(*graph, graph_name.GetString(), 7);
             // (optional). Save model
             aclgrphSaveModel(graph_name.GetString(), *om_model);
 #endif