VPI CUDA, opencv GPU Mat

Robotics & Edge Computing Jetson Systems Jetson AGX Orin
1

Continuing the discussion from VPI CUDA,OpenCV Interoperablity:

With VPI version 3.2, I can’t seem to find the vpiImageCreateCUDAMemWrapper in documentation, does it still exist ? if not, how to create a VPIImage from an OpenCV GPU Mat.
Alternatively, I work with Deepstream, is there a better way to create a VPIImage from the NvBufSurface buffer
currently i’m doing this :

    in_surface = (NvBufSurface *) in_map_info.data;

    if (NvBufSurfaceMapEglImage(in_surface, 0) != 0) {
        goto error;
    }
#ifdef __aarch64__
    CUresult status;
    CUeglFrame in_eglFrame;
    CUgraphicsResource pInResource;
    pInResource = nullptr;

    cudaFree(0);

    status = cuGraphicsEGLRegisterImage(&pInResource, in_surface->surfaceList[0].mappedAddr.eglImage, CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE);
    status = cuGraphicsResourceGetMappedEglFrame(&in_eglFrame, pInResource, 0, 0);

    status = cuCtxSynchronize();
    img_gpu = new cv::cuda::GpuMat(in_eglFrame.height, in_eglFrame.width, CV_8UC4, in_eglFrame.frame.pPitch[0]);
#else
    img_gpu = new cv::cuda::GpuMat(in_surface->surfaceList[0].height, in_surface->surfaceList[0].width, CV_8UC4, in_surface->surfaceList[0].dataPtr);
#endif

Thank you

4

Hi,

The API changes to :

https://docs.nvidia.com/vpi/group__VPI__Image.html#ga3e7cf2520dd568a7e7a9a6876ea7995c

Thanks.