Merge pull request #24709 from vpisarev:winograd_mode

Try to enable Winograd by default in FP32 mode and disable it by default in FP16 mode #24709

Hopefully, it will resolve regressions since 4.8.1 (see also https://github.com/opencv/opencv/pull/24587)

### 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

modules/dnn/include/opencv2/dnn/all_layers.hpp

@@ -291,7 +291,7 @@ CV__DNN_INLINE_NS_BEGIN
         static Ptr<BaseConvolutionLayer> create(const LayerParams& params);
         bool fusedActivation = false;
         bool fusedAdd = false;
-        bool useWinograd = false; // Flag whether to use Winograd to speed up 3x3 convolution.
+        bool useWinograd = true; // Flag whether to use Winograd to speed up 3x3 convolution.
     };
 
     class CV_EXPORTS ConvolutionLayerInt8 : public BaseConvolutionLayer

modules/dnn/src/layers/layers_common.cpp

@@ -195,7 +195,7 @@ void getConvolutionKernelParams(const LayerParams &params, std::vector<size_t>&
     util::getStrideAndPadding(params, pads_begin, pads_end, strides, padMode, kernel.size());
     util::getParameter(params, "dilation", "dilation", dilations, true, std::vector<size_t>(kernel.size(), 1));
     util::getParameter(params, "adj", "adj", adjust_pads, true, std::vector<size_t>(kernel.size(), 0));
-    useWinograd = params.get<bool>("use_winograd", true);
+    useWinograd = params.get<bool>("use_winograd", useWinograd);
 
     for (int i = 0; i < dilations.size(); i++)
         CV_Assert(dilations[i] > 0);

modules/dnn/src/net_impl_backend.cpp

@@ -251,6 +251,14 @@ void Net::Impl::setPreferableTarget(int targetId)
 #endif
 
         clear();
+
+        if (targetId == DNN_TARGET_CPU_FP16)
+        {
+            if (useWinograd) {
+                CV_LOG_INFO(NULL, "DNN: DNN_TARGET_CPU_FP16 is set => Winograd convolution is disabled by default to preserve accuracy. If needed, enable it explicitly using enableWinograd(true).");
+                enableWinograd(false);
+            }
+        }
     }
 }