Jetson docker image extension

Hi, i try to extend base docker container for jetson tx2 and try to use recomendation from this answer, but i have error:

dpkg: dependency problems prevent configuration of libnvinfer6:
 libnvinfer6 depends on cuda-cublas-10-0; however:
  Package cuda-cublas-10-0 is not installed.
 libnvinfer6 depends on cuda-cudart-10-0; however:
  Package cuda-cudart-10-0 is not installed.

dpkg: error processing package libnvinfer6 (--install):
 dependency problems - leaving unconfigured

I’m can not find cudart and cublas deb files.

How i can extend base docker images to using cuda and cudnn for install some libraries like opencv for cuda, dlib?

Hi,

Here is a docker image with CUDA, cuDNN and TensorRT installed.
So you don’t need to install them manually.

https://ngc.nvidia.com/catalog/containers/nvidia:deepstream-l4t

Thanks.

Thank you for answer but cublas and other dependences not to be found when i try to install opencv.

pthread cudart_static -lpthread dl rt CUDA_nppc_LIBRARY-NOTFOUND CUDA_nppial_LIBRARY-NOTFOUND CUDA_nppicc_LIBRARY-NOTFOUND CUDA_nppicom_LIBRARY-NOTFOUND CUDA_nppidei_LIBRARY-NOTFOUND CUDA_nppif_LIBRARY-NOTFOUND CUDA_nppig_LIBRARY-NOTFOUND CUDA_nppim_LIBRARY-NOTFOUND CUDA_nppist_LIBRARY-NOTFOUND CUDA_nppisu_LIBRARY-NOTFOUND CUDA_nppitc_LIBRARY-NOTFOUND CUDA_npps_LIBRARY-NOTFOUND CUDA_cublas_LIBRARY-NOTFOUND CUDA_cufft_LIBRARY-NOTFOUND -L/usr/local/cuda-10.0/lib64 -L/usr/lib/aarch64-linux-gnu

I have all of them in /usr/local/cuda-10.0/lib64 on my host jetson, but have not on image.

I think what i’m use false path to cuda (cudnn) libs (i’m use JEP for instalation)

After trying FROM nvcr.io/nvidia/deepstream-l4t:4.0.2-19.12-samples have same error.
Not find cudnn too.

dpkg: dependency problems prevent configuration of libnvinfer6:
 libnvinfer6 depends on libcudnn7; however:
  Package libcudnn7 is not installed.
 libnvinfer6 depends on cuda-cublas-10-0; however:
  Package cuda-cublas-10-0 is not installed.
 libnvinfer6 depends on cuda-cudart-10-0; however:
  Package cuda-cudart-10-0 is not installed.

Hi,

TensorRT (nvinfer) is pre-installed in the image.
So you don’t need to manually install it again.

Do we miss anything?

Thanks.

Are the installation paths of cudnn and cuda different with host?
The packages I need cannot be found in the directories in which they are installed on the host.

When i try install libnvinfer from jetpack downloads its write me:

dpkg: dependency problems prevent configuration of libnvinfer6:
 libnvinfer6 depends on libcudnn7; however:
  Package libcudnn7 is not installed.
 libnvinfer6 depends on cuda-cublas-10-0; however:
  Package cuda-cublas-10-0 is not installed.
 libnvinfer6 depends on cuda-cudart-10-0; however:
  Package cuda-cudart-10-0 is not installed.

It says what cudnn not installed? Or maybe it installed in different path with host system?

Hi,

cuDNN is installed in /usr/lib/aarch64-linux-gnu/ on a Jetson device.

Thanks.

This image can use with Jetson ?
And is it possible to use OpenCV with CUDA ?
If it dowsn’t include OpenCV, I consider using nano_build_opencv.
And nano_build_opencv can use for Docker in Xavier NX ?

Hi hkada,

Please open a new topic for your issue. Thanks