SDK Manager - OpenCV Cuda support

Using latest SDK manager to install CUDA and OpenCV, but OpenCV does not appear to have CUDA support. Am I doing something wrong??? SURELY SDK manager should install opencv with cuda support???

CUDA is 10.0
OpenCV is 3.3.1

but from cv::getBuildInformation…

Other third-party libraries:
Use Intel IPP: NO
Use Intel IPP IW: NO
Use VA: NO
Use Intel VA-API/OpenCL: NO
Use Lapack: NO
Use Eigen: YES (ver 3.3.4)
Use Cuda: NO
Use OpenCL: NO
Use OpenVX: NO
Use custom HAL: YES (carotene (ver 0.0.1))

1 Like

Hi peter.hawkin,

Please try with the recently release JetPack 4.2.1 through SDK manager.

Thanks

I have just tried this and it still does not compile opencv code which uses the cuda namespace.

Why did you make this suggestion - do you know for user that the opencv version installed by nvidiasdk is now compiled with CUDA?

Peter

1 Like

Hi,

Our official OpenCV package does support CUDA. The release in each JetPack does.
But there are some third-party libraries are not enabled inside the package due to license issue, ex. GStreamer.

May I know the exact issue you meet in detail?

Thanks.

If the standard install of OpenCV from SDK manager installs a cuda version of OpenCV, then why does getBuildInformation say that cuda is not supported? (see my original post).

I found a few links on the web supporting my theory that SDKmanager does not install a cuda-build of OpenCV and some workarounds.

To answer your question, as a first step it would be good if cv::getBuildInformation showed Use Cuda: YES

thanks,

Peter

1 Like

according to the thread 1063091:

https://devtalk.nvidia.com/default/topic/1063091/jetson-agx-xavier/opencv-cuda-/

sdkmanager installed opencv package does not support CUDA.

Please try to build the opencv + cuda from source.

https://github.com/AastaNV/JEP/tree/master/script

1 Like

I encountered the same issue while flashing through JetPack 4.3. Moreover and specifically the JetPack installed OpenCV is for Python 2.* rather than Python 3.* which soon is deprecating.

Same here!

  • NVIDIA Jetson AGX Xavier
    • Jetpack 4.3 [L4T 32.3.1]
    • CUDA GPU architecture 7.2
    • NV Fan Mode:quiet - Type: NV Power Mode: MAXN
  • Libraries:
    • CUDA 10.0.326
    • cuDNN 7.6.3.28-1+cuda10.0
    • TensorRT 6.0.1.10-1+cuda10.0
    • Visionworks 1.6.0.500n
    • OpenCV NOT_INSTALLED compiled CUDA: NO
  • Jetson Performance: active

It definitely should install cuda enabled OpenCV.

If your OpenCV does not have CUDA support,
you should build to install it from source with -D WITH_CUDA=ON flag.
…

-D WITH_CUDA=ON
-D ENABLE_FAST_MATH=1
-D CUDA_FAST_MATH=1
-D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.0
-D WITH_CUBLAS=1
…