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!

Alert: Beginning Tuesday, June 25th, we will be freezing this site and migrating the content and forums to our new home at https://forums.intel.com. Check it out now!

RuntimeError: Bad magic number when installing NC API

The ./convert_models.sh step spits only that error. Python version is 3.5.3.

Thanks,

Tom

Comments

  • 12 Comments sorted by Votes Date Added
  • Intel Graphics Driver causes my Windows 10 screen to frequently go black.

  • Hi @fortiss,
    I tried reproducing your issue on a couple existing systems (on which NCS worked well), and also made a fresh install on a clean Ubuntu 16.04 machine; all of these systems have python 3.5.2. The toolkit/setup.sh script installs python3, but it doesn't really specify a sub-version, I see that the package version for Ubuntu 16.04 is 3.5.1-3 (which is actually 3.5.2). Perhaps you manually installed python 3.5.3 on your machine? If so, can you please uninstall/purge it, run sudo apt-get install python3 python3-pip, and then run ncapi/setup.sh? (you probably don't have to rerun toolkit/setup.sh)

    Excerpt from toolkit/setup.sh:
    $SUDO_PREFIX apt-get $APT_QUIET install -y unzip coreutils curl git python3 python3-pip

    PS: Let me know if above suggestion works, I am curious to know if the problem is due to the difference in subversion.

  • It looks like python made a breaking change in 3.5.3 and updated the magic number. http://bugs.python.org/issue27286

  • Hi @"Ashwin Vijayakumar",

    I trying to install sdk on Ubuntu 17.04 and have a similar problem because it uses python 3.5.3. It seems it will be easy to fix api build if I can rebuild mvNCCompile.pyc from source.

  • @albor thanks for trying out the SDK on 17.04, I was building a fresh machine with 17.04 so I could test. The API and toolkit currently aren't OpenSouce, so can you please move to a 16.04 machine with 3.5.2?

  • @AshwinVijayakumar,

    I downgraded to 16.04 and was able to run samples :). Unfortunately on 17.04 as far as I understand graphviz python wrapper is not "binding" with graphviz properly out of the box. That prevents movidius sdk setup even if magic number problem resolved.

  • python3 ./mvNCCompile.pyc data/lenet8.prototxt -s 12
    RuntimeError: Bad magic number in .pyc file
    Makefile:21: recipe for target 'example00' failed
    make: *** [example00] Error 1
    

    With Python 3.4.2 on BunsenLabs (Debian Jessie)

  • @BQBQ Probably the difference between BunsenLabs and Ubuntu is causing this. Please try to use Ubuntu 16.04 (and Python 3.5.2 by default on it) as recommended in "Getting Started Guide" which is proven to run examples correctly.

  • Would it be possible to get the non-cythoned version of the mvNCCompile script? It's possible to navigate around most other problems on non-Ubuntu systems (such as Debian / raspbian / ubilinux, etc), but this Python sub-version dependency essentially requires one to install Ubuntu in a chroot, which complicates deployment work quite a bit.

  • @gekko Your concern is a valid and the SDK's dependency on Python 3.5.2 is an issue that we know about and one that we are looking into.

  • So I would comment here:
    https://ncsforum.movidius.com/discussion/242/bad-magic-number-in-pyc-file
    , but also getting the error while building the NCAPI via shell script.

    Saving to: ‘../images/cat.jpg’
    
    cat.jpg             100%[===================>] 137.10K  --.-KB/s    in 0.03s   
    
    Last-modified header missing -- time-stamps turned off.
    2017-10-10 13:14:00 (4.90 MB/s) - ‘../images/cat.jpg’ saved [140391/140391]
    
    ../../bin
    RuntimeError: Bad magic number in .pyc file
    RuntimeError: Bad magic number in .pyc file
    RuntimeError: Bad magic number in .pyc file
    RuntimeError: Bad magic number in .pyc file
    RuntimeError: Bad magic number in .pyc file
    

    Thoughts?

  • @karl What python version are you using? Assuming this is with the 1.07.10 release, you can also give the newest release a try: https://github.com/movidius/ncsdk/releases/tag/v1.09.00.06 and for more information: https://ncsforum.movidius.com/discussion/98/latest-version-movidius-neural-compute-sdk#latest

Sign In or Register to comment.