CUDA backend for the opencv_dnn
Does anybody know if I can somewhere find the opencv_dnn module with CUDA support? A master branch supports only MKL and OpenCL backends but both too slow compare to the Caffe with CUDA: in my particular task average forward propagation time in Caffe with CUDA takes about 15 ms whereas in opencv_dnn with MKL 350 ms and with OpenCL 280 ms.
You need to built it from scratch. There are no prepacked modules with CUDA support.
I've asked not about how to build, but about existence of CUDA backend for the opencv dnn, maybe someone already works on it?
am facing the same problem but on tensorflow, however Ican't make it forward propagate on opencl, how could you use opencl to increase the inference time?