opencv
diff --git a/‎modules/dnn/src/cuda/eltwise_ops.cu
Copy file name to clipboard
+7Lines changed: 7 additions & 0 deletions b/‎modules/dnn/src/cuda/eltwise_ops.cu
Copy file name to clipboard
+7Lines changed: 7 additions & 0 deletions
diff --git a/‎modules/dnn/src/cuda/functors.hpp
Copy file name to clipboard
+15Lines changed: 15 additions & 0 deletions b/‎modules/dnn/src/cuda/functors.hpp
Copy file name to clipboard
+15Lines changed: 15 additions & 0 deletions
diff --git a/‎modules/dnn/src/cuda4dnn/kernels/eltwise_ops.hpp
Copy file name to clipboard
+3Lines changed: 3 additions & 0 deletions b/‎modules/dnn/src/cuda4dnn/kernels/eltwise_ops.hpp
Copy file name to clipboard
+3Lines changed: 3 additions & 0 deletions
diff --git a/‎modules/dnn/src/cuda4dnn/primitives/eltwise.hpp
Copy file name to clipboard
+8-3Lines changed: 8 additions & 3 deletions b/‎modules/dnn/src/cuda4dnn/primitives/eltwise.hpp
Copy file name to clipboard
+8-3Lines changed: 8 additions & 3 deletions
@@ -350,6 +350,11 @@ void eltwise_fmod_2(const Stream& stream, TensorSpan<T> output, TensorView<T> x,
     eltwise_op<T, FModFunctor<T>>(stream, output, x, y);
 }
 
+template <class T>
+void eltwise_pow_2(const Stream& stream, TensorSpan<T> output, TensorView<T> x, TensorView<T> y) {
+    eltwise_op<T, PowFunctor<T>>(stream, output, x, y);
+}
+
 #if !defined(__CUDA_ARCH__) || (__CUDA_ARCH__ >= 530)
     template void eltwise_mod_2(const Stream& stream, TensorSpan<__half> output, TensorView<__half> x, TensorView<__half> y);
     template void eltwise_fmod_2(const Stream& stream, TensorSpan<__half> output, TensorView<__half> x, TensorView<__half> y);
@@ -360,6 +365,7 @@ void eltwise_fmod_2(const Stream& stream, TensorSpan<T> output, TensorView<T> x,
     template void eltwise_sum_2(const Stream& stream, TensorSpan<__half> output, TensorView<__half> x, TensorView<__half> y);
     template void eltwise_max_2(const Stream& stream, TensorSpan<__half> output, TensorView<__half> x, TensorView<__half> y);
     template void eltwise_min_2(const Stream& stream, TensorSpan<__half> output, TensorView<__half> x, TensorView<__half> y);
+    template void eltwise_pow_2(const Stream& stream, TensorSpan<__half> output, TensorView<__half> x, TensorView<__half> y);
 #endif
     template void eltwise_mod_2(const Stream& stream, TensorSpan<float> output, TensorView<float> x, TensorView<float> y);
     template void eltwise_fmod_2(const Stream& stream, TensorSpan<float> output, TensorView<float> x, TensorView<float> y);
@@ -370,5 +376,6 @@ void eltwise_fmod_2(const Stream& stream, TensorSpan<T> output, TensorView<T> x,
     template void eltwise_sum_2(const Stream& stream, TensorSpan<float> output, TensorView<float> x, TensorView<float> y);
     template void eltwise_max_2(const Stream& stream, TensorSpan<float> output, TensorView<float> x, TensorView<float> y);
     template void eltwise_min_2(const Stream& stream, TensorSpan<float> output, TensorView<float> x, TensorView<float> y);
+    template void eltwise_pow_2(const Stream& stream, TensorSpan<float> output, TensorView<float> x, TensorView<float> y);
 
 }}}} /* namespace cv::dnn::cuda4dnn::kernels */
@@ -833,6 +833,21 @@ struct FModFunctor {
     }
 };
 
+template <class T>
+struct PowFunctor {
+    struct Params {
+        CUDA4DNN_HOST_DEVICE Params() {}
+    };
+
+    CUDA4DNN_DEVICE PowFunctor() { }
+    CUDA4DNN_DEVICE PowFunctor(const Params& params) { }
+
+    CUDA4DNN_DEVICE T operator()(T x, T y) {
+        using csl::device::pow;
+        return pow(x, y);
+    }
+};
+
 }}}} /* namespace cv::dnn::cuda4dnn::kernels */
 
 #endif /* OPENCV_DNN_SRC_CUDA_FUNCTORS_HPP */
@@ -39,6 +39,9 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace kernels {
     template <class T>
     void eltwise_fmod_2(const csl::Stream& stream, csl::TensorSpan<T> output, csl::TensorView<T> x, csl::TensorView<T> y);
 
+    template <class T>
+    void eltwise_pow_2(const csl::Stream& stream, csl::TensorSpan<T> output, csl::TensorView<T> x, csl::TensorView<T> y);
+
 }}}} /* namespace cv::dnn::cuda4dnn::kernels */
 
 #endif /* OPENCV_DNN_SRC_CUDA4DNN_KERNELS_ELTWISE_OPS_HPP */
@@ -30,6 +30,7 @@ namespace cv { namespace dnn { namespace cuda4dnn {
         SUB,
         MOD,
         FMOD,
+        POW,
     };
 
     class EltwiseOpBase : public CUDABackendNode {
@@ -62,7 +63,6 @@ namespace cv { namespace dnn { namespace cuda4dnn {
             const std::vector<cv::Ptr<BackendWrapper>>& outputs,
             csl::Workspace& workspace) override
         {
-            CV_Assert(inputs.size() >= 2);
             CV_Assert(outputs.size() == 1);
 
             CV_Assert(coeffs.size() == 0 || op == EltwiseOpType::SUM);
@@ -94,9 +94,13 @@ namespace cv { namespace dnn { namespace cuda4dnn {
                 case EltwiseOpType::SUB: kernels::eltwise_sub_2<T>(stream, output, input_x, input_y); break;
                 case EltwiseOpType::MOD: kernels::eltwise_mod_2<T>(stream, output, input_x, input_y); break;
                 case EltwiseOpType::FMOD: kernels::eltwise_fmod_2<T>(stream, output, input_x, input_y); break;
+                case EltwiseOpType::POW: kernels::eltwise_pow_2<T>(stream, output, input_x, input_y); break;
                 }
-            }
-            else
+            } else if (inputs.size() == 1) {
+                auto input_wrapper_0 = inputs[0].dynamicCast<wrapper_type>();
+                auto input_0 = input_wrapper_0->getView();
+                csl::tensor_ops::copy(stream, output, input_0);
+            } else
             {
                 auto input_wrapper_0 = inputs[0].dynamicCast<wrapper_type>();
                 auto input_0 = input_wrapper_0->getView();
@@ -128,6 +132,7 @@ namespace cv { namespace dnn { namespace cuda4dnn {
                     case EltwiseOpType::SUB: kernels::eltwise_sub_2<T>(stream, output, output, input); break;
                     case EltwiseOpType::MOD: kernels::eltwise_mod_2<T>(stream, output, output, input); break;
                     case EltwiseOpType::FMOD: kernels::eltwise_fmod_2<T>(stream, output, output, input); break;
+                    case EltwiseOpType::POW: kernels::eltwise_pow_2<T>(stream, output, output, input); break;
                     }
                 }
             }