Merge pull request #16220 from YashasSamaga:cuda4dnn-roi-pooling-test_fix-optim

modules/dnn/src/cuda4dnn/kernels/roi_pooling.hpp

@@ -12,7 +12,7 @@
 namespace cv { namespace dnn { namespace cuda4dnn { namespace kernels {
 
     template <class T>
-    void roi_pooling(const csl::Stream& stream, csl::TensorSpan<T> output, csl::TensorView<T> input, csl::View<T> rois, T spatial_scale);
+    void roi_pooling(const csl::Stream& stream, csl::TensorSpan<T> output, csl::TensorView<T> input, csl::View<T> rois, float spatial_scale);
 
 }}}} /* namespace cv::dnn::cuda4dnn::kernels */
 

modules/dnn/test/test_layers.cpp

@@ -600,6 +600,11 @@ TEST_P(Test_Caffe_layers, ROIPooling_Accuracy)
 
     double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1e-3 : 1e-5;
     double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1e-3 : 1e-4;
+    if (target == DNN_TARGET_CUDA_FP16)
+    {
+        l1 = 2e-4;
+        lInf = 9e-4;
+    }
     normAssert(out, ref, "", l1, lInf);
 }