VisionWorks install without JetPack

Hello people!

I have a simple question - is there a way to get the .deb-package of the VisionWorks SDK without the JetPack installer?

I think you will hate me at this point: But i use Arch Linux and not Ubuntu at work.

Since installing debian packages on Arch is no problem, i am able to install the SDK-Manager for the JetPack (https://developer.nvidia.com/embedded/jetpack),
but it refuses to download anything - since it is not running on an Ubuntu machine.

It would be cool, if there is any option.
OpenCV, Cuda and the NVidia SDK are running on my machine without any problems - so i don’t see why VisionWorks should have a problem.
Also I am fine by installing without a fancy gui :)

Thank you for your help!

Hi,

Would you mind to try our command line mode of sdkmanager.
It might be work on your system.

Here is the document cli for your reference:
https://docs.nvidia.com/sdk-manager/sdkm-command-line-install/index.html

Thanks.

A colleague of mine wanted to use Debian and found ways around the artificial limits imposed in SDK Manager. His notes on the situation may help you:

“Once installed, it will check your lsb_release output and stop you if it is not the same as what Nvidia supports. You will see an error that says something like “OS Linux not supported” at the first step of downloading anything. If you throw something like this into your path via ~/.local/bin/ or whatever your preference; we can fake to sdkmanager that we are Ubuntu in a “safer” way. It only looks at the lsb_release -r output as of August 16 2019.”

#!/bin/bash
# Fake lsb_release. I am Ubuntu.

if (pgrep -x sdkmanager >/dev/null) && [[ ! -z $1 && $1=="-r" ]]; then
	echo "Release:	18.04"
	exit 0
fi

/usr/bin/lsb_release $@

Hi AastaLLL,
yes, i tried this, but it also complains about me not using Ubuntu 18.04 - just over the command line instead of a gui.

Hm, this is a cool solution; but works only for debian, because ubuntu and debian have the same package managing system (dpkg ; ye, Ubuntu uses apt, but this is basically just a wrapper around dpkg as far as i am informed).

But i made some progress:

I took an Ubuntu, and used the SDK Manager to download the *.deb packages,
copied them to my Arch and was partly able to convert them into Arch Packages.
The Problem is, that the debian packages are nested - so one package contains three other ones and therefore the standard tool to convert debian packages was not able to handle this properly.
Long story short, I ripped them all apart and I think the installation is almost complete?
I have the same files under /usr/libs and /usr/include if i compare this to the Ubuntu system.
Nevertheless - if i want to compile the Demos of the VisionWorks SDK, i get an error

Package cudart-10.0 was not found in the pkg-config search path.
Perhaps you should add the directory containing `cudart-10.0.pc'
to the PKG_CONFIG_PATH environment variable

Which makes sense, because i can’t find a cudart-10.0.pc in my package configuration.
cudart is installed though (at least i have all *.so files of it?)

Does anyone know, where this cudart.pc file usually lies or what is in it?

Maybe this is the only problem and can be fixed :)

Thanks for your time and help so far!

D3[ L✔] d3-jetson-bsp {}$ sudo find / -name 'cudart*.pc'
/media/growe/65b870bd-0aa7-4325-b79f-1d08c1c66ed1/usr/lib/pkgconfig/cudart-10.0.pc
/usr/lib/pkgconfig/cudart-10.0.pc

The mounted media is a Xavier system.

The pc file is contained in the cuda-cudart-dev package:

D3[ L✔] d3-jetson-bsp {}$ dpkg -S /usr/lib/pkgconfig/cudart-10.0.pc 
cuda-cudart-dev-10-0: /usr/lib/pkgconfig/cudart-10.0.pc

The contents of the pkg-config file are the same on my host and target:

nvidia@nvidia-desktop:~$ cat /usr/lib/pkgconfig/cudart-10.0.pc 
cudaroot=/usr/local/cuda-10.0
libdir=${cudaroot}/lib64
includedir=${cudaroot}/include

Name: cudart
Description: CUDA Runtime Library
Version: 10.0
Libs: -L${libdir} -lcudart
Cflags: -I${includedir}

Thank you for your work!

I created the file and was able to compile the examples,
but i can not execude the binaries - because of this:

./nvx_demo_motion_estimation: /opt/cuda/lib64/libcudart.so.10.0: version `libcudart.so.10.0' not found (required by /usr/lib64/libvisionworks.so)

I am giving up now (guess it is not meant to be) and will use a VM with Ubuntu.
But it’s sad, that no other Platform than an old Ubuntu Version is supported for building this Software…

Thanks for your help all!