frame

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Installation fails with errors

Hi guys,

I've been trying to install the SDK on Ubuntu 16.04 but the process seems very flawed. Notably:
(1) You can't install without root as it tries to do a whole bunch of operations on /opt and gets permission denied
(2) Installation as root fails with "Installation failed. Error on line 303" when trying to install tensorflow. I saw in another thread this is a known issue
(3) In fact I have tensorflow installed but it keeps throwing "command not found" errors.

Kind of a bummer for somebody excited to get this product. Any advice out there? Maybe I can get a docker image with all this stuff already correct or something? I spoke to another friend of mine and he said he had to jump through incredible hoops to get this installed.

Comments

  • 7 Comments sorted by Votes Date Added
  • I ran as root and not as root. Running as root actually seemed to make things worse. My basic problem is I am starting from bare Ubuntu 16.04 and the make install command installs a ton of dependencies. That might keep things simple but I would much rather install all the dependencies by hand so I can see what is going on.

    As it is, I also have an error installing tensorflow. See below. I don't have tensorflow installed. In fact I have nothing installed.

    Movidius Neural Compute Toolkit Setup.
    Checking OS and version...
    Installing on Ubuntu 16.04

    Installer NCSDK version: 1.09.00.04
    ...

    Checking whether tensorflow is installed...
    looking for tensorflow CPU version...
    You are using pip version 8.1.1, however version 9.0.1 is available.
    You should consider upgrading via the 'pip install --upgrade pip' command.
    looking for tensorflow GPU version...
    You are using pip version 8.1.1, however version 9.0.1 is available.
    You should consider upgrading via the 'pip install --upgrade pip' command.
    Couldn't find a supported tensorflow version, installing tensorflow v1.3.0
    Collecting tensorflow==1.3.0
    Could not find a version that satisfies the requirement tensorflow==1.3.0 (from versions: )
    No matching distribution found for tensorflow==1.3.0
    You are using pip version 8.1.1, however version 9.0.1 is available.
    You should consider upgrading via the 'pip install --upgrade pip' command.
    Installation failed. Error on line 303

  • @Yakumo01, @EricPeterson, we are aware of the installation issues w.r.t root and are working on a fix for the next release. Until then, you can choose to install in a non root account or follow @pisymbol 's workaround - https://ncsforum.movidius.com/discussion/comment/707/#Comment_707

  • That might keep things simple but I would much rather install all the dependencies by hand so I can see what is going on

    @EricPeterson, you are right about keeping the installation simple, that's why we have abstracted everything with a simple shell script. If you choose to install the dependencies manually, a good starting point would be install-ncsdk.sh, requirements.txt and requirements-apt.txt found in NCSDK-XXXX.tar.gz.

    To manually download the tarball, just follow this path

    1. Look for ncsdk_redirector.txt in https://github.com/movidius/ncsdk/blob/master/install.sh
    2. Look for NCSDK-XXXXX.tar.gz's download link in ncsdk_redirector.txt
  • Thanks for the link Ashwin, I am trying the install script fix now.

    I tried installing everything by hand but failed. What stopped me was trying to install Bazel by bootstrapping. Tensorflow requires Bazel and Bazel provides binary packages for intel. But I didn't see any package for ARM. So I tried bootstrapping Bazel which runs java for hours and then fails on a "resource" error. I suspect the raspberry pi is insufficient to bootstrap Bazel or things like that.

    As a compromise I would suggest providing a script to install dependencies like tensorflow which depends on Bazel and probably a lot of other things. Then perform the ncsdk build in a separate script once the dependencies are done. The first script can do things with sudo. The second script should not use sudo. That's a key difference. Running large portions of scripts with sudo is a flawed methodology.

  • Hi, thanks for the feedback. I did eventually get it to install but for the life of me I can't recall how :)

  • I ran the initial install.sh script until it failed with the error message above (ends with 303). Then I edited the install-ncsdk.sh file in /opt/movidius as suggested in the 707 comment linked above. I also added -x to the bash command at the top of the script. Then executed that script. It exits quickly with these last lines:

    +++ unset _def _dir _file _nosp _minusdd _minusdf
    ++ for i in '$(LC_ALL=C command ls "$BASH_COMPLETION_COMPAT_DIR")'
    ++ i=/etc/bash_completion.d/youtube-dl.bash-completion
    ++ [[ youtube-dl.bash-completion != @(@(#*#|*@(~|.@(bak|orig|rej|swp|dpkg*|rpm@(orig|new|save))))|Makefile*|@(acroread.sh)
    ) ]]
    ++ [[ -f /etc/bash_completion.d/youtube-dl.bash-completion ]]
    ++ [[ -r /etc/bash_completion.d/youtube-dl.bash-completion ]]
    ++ . /etc/bash_completion.d/youtube-dl.bash-completion
    +++ complete -F __youtube_dl youtube-dl
    ++ unset i _blacklist_glob
    ++ [[ /usr/share/bash-completion/bash_completion != \/\h\o\m\e\/\e\r\i\c/.bash_completion ]]
    ++ [[ -r /home/eric/.bash_completion ]]
    ++ unset -f have
    ++ unset have
    ++ set +v
    ++ unset BASH_COMPLETION_ORIGINAL_V_VALUE

  • Hi, you have to run installation only with "make install" on your user account without sudo. On some point install script may ask you for your password. Running installation as root fails and that was mentioned in other posts.

Sign In or Register to comment.