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!

Fresh Install on Pi taking forever

Hi,
I messed up my Pi image and have to rebuild. So I downloaded stretch and installed.
Is anyone else having issues with the install of NCSDK ?
The Git Hub was not there so I downloaded the tar file unzipped into the workspace/ncsdk folder and ran
"make install" command. Its been running over 16 hours trying to build some OpenGL libraries on a Raspberry Pi 3
and this morning I noticed some compiler error but it was still compiling.
Anyone else having issues with this?

Comments

  • 13 Comments sorted by Votes Date Added
  • edited September 2018 Vote Up0Vote Down

    @chicagobob123
    I hope it's a useful reference.
    Extension of SWAP area and specification of option "-j1"(MAKE_NJOBS=1) of make command is required.
    It will finish in 2 hours ~ 3 hours.
    https://github.com/PINTO0309/MobileNet-SSD-RealSense#work-with-raspberrypi3-or-pc--ubuntu1604--raspberrypi--ubuntu-mate

  • Thanks PINTO I will increase the swap size.

  • @chicagobob123, on a RPi the full installation (make install) will build OpenCV from source, but I am not aware of OpenGL build. If you can live with just inference capabilities on your RPi (compiling model on your laptop/desktop, and then transferring the graph to RPi via SSH or USB), just install the API framework, it wouldn't take more than 10 mins (including dependencies). See instructions here - https://movidius.github.io/blog/ncs-apps-on-rpi/

  • Thank you Ashwin. So sorry, I was posting from memory and openCV was what was building. It errored out on compile from make install.
    I increased the swapfile size and going to try again.

  • PINTO I have been following your install instructions.
    So far it installed but when tried to run the hello .py it froze.
    I did not have time last night to try again. Let you know when I figure out whats going on. In all the failed attempts maybe something is just installed wrong. really wanted this out of my hair to move forward. I was going to post my version of the face detector but got nothing to post with.
    On the raspberry pi forum, I was told that the Pi Camera should be faster.

  • edited September 2018 Vote Up0Vote Down

    @chicagobob123
    In my experience, the graph converted from Tensorflow is very unstable.
    Often it froze and did not work properly.
    Since NCSDK 's API is still unstable, I gave up converting from Tensorflow a long time ago.
    If you are faithfully following the procedure I suggested, there is probably no problem with the installation.
    I think there is a problem with the graph file.
    Can you post a line of program that causes freezing?
    Is it possible to tell me the URL of raspberry pi forum?

  • Will check out the rest tonight. Here is the pi forum where I posted my request.
    https://www.raspberrypi.org/forums/viewtopic.php?f=43&t=221425&p=1359515#p1359515

  • edited September 2018 Vote Up0Vote Down

    @chicagobob123
    It is a very interesting topic.
    As long as you use openCV, the performance of cv2.waitKey() becomes a bottleneck.
    cv2.waitKey(1) = 15ms wait
    cv2.waitKey(16) = 30ms wait
    cv2.waitKey(31) = 45ms wait

    【Japanese Article】http://13mzawa2.hateblo.jp/entry/2015/12/28/180021

    The theoretical transfer rate of USB 2.0 is 270 FPS at 320x240.
    Calculations may be wrong,
    Size of 1 image:
    320x240x24/8=230,400byte
    230,400byte/1024=225KB
    USB2.0:
    60MB/secx1024=61440KB
    61440KB/225KB=273Frame/sec

    By the way, Specification of USB Camera [PlayStationEye] that I am using.

    Resolution Effective frame rate of USB Camera
    【320x240】187.000, 150.000, 137.000, 125.000, 100.000, 75.000, 60.000, 50.000, 37.000, 30.000
    【640x480】60.000, 50.000, 40.000, 30.000, 15.000

    If you do not need cv2.imshow(), cv2.waitKey() will be unnecessary.
    I will try to verify with PlayStationEye when I get home after work today.

  • edited September 2018 Vote Up0Vote Down

    @chicagobob123
    Results different from expectations came out.

    PlayStationEye(USB2.0) + 320x240 + OpenCV's VFPV3=ON + No image processing
    Available cv2.imshow() + cv2.waitKey(1) = about 150 FPS
    No cv2.imshow() + cv2.waitKey(1) = about 150 FPS
    6.7ms / 1frame

    PlayStationEye(USB2.0) + 640x480 + OpenCV's VFPV3=ON + No image processing
    Available cv2.imshow() + cv2.waitKey(1) = about 60 FPS
    No cv2.imshow() + cv2.waitKey(1) = about 60 FPS
    16.7ms / 1frame

    The internal operation of OpenCV seems to be improved.
    If you do some kind of image processing, it will be quite late.

  • I got the oddest results from running hello_ncs_py
    ??? Can anyone help with this?

    making run
    python3 hello_ncs.py;
    D: [ 0] ncDeviceCreate:307 ncDeviceCreate index 0

    D: [ 0] ncDeviceCreate:307 ncDeviceCreate index 1

    D: [ 0] ncDeviceOpen:501 File path /usr/local/lib/mvnc/MvNCAPI-ma2450.mvcmd

    I: [ 0] ncDeviceOpen:507 ncDeviceOpen() XLinkBootRemote returned success 0

    I: [ 0] ncDeviceOpen:536 XLinkConnect done - link Id 0

    D: [ 0] ncDeviceOpen:550 done

    I: [ 0] ncDeviceOpen:552 Booted 1.3-ma2450 -> VSC

    I: [ 0] getDevAttributes:373 Device attributes

    I: [ 0] getDevAttributes:376 Device FW version: 2.4.2450.f0

    I: [ 0] getDevAttributes:378 mvTensorVersion 2.4

    I: [ 0] getDevAttributes:379 Maximum graphs: 10

    I: [ 0] getDevAttributes:380 Maximum fifos: 20

    I: [ 0] getDevAttributes:382 Maximum graph option class: 1

    I: [ 0] getDevAttributes:384 Maximum device option class: 1

    I: [ 0] getDevAttributes:385 Device memory capacity: 522073264

    Hello NCS! Device opened normally.
    I: [ 0] ncDeviceClose:742 closing device

  • edited September 2018 Vote Up0Vote Down

    @chicagobob123
    It does not cause an error, it seems to be working normally.

    What will happen to the following?
    hello_ncs_api2.py

    https://github.com/movidius/ncappzoo/blob/ncsdk2/apps/hello_ncs_py/hello_ncs_api2.py

  • I have not downloaded the appzoo yet Just wanted to get this part working.
    I will have it today.
    The CPP version of hello reponds as expected

  • WOW, I am finally working again. I have not built the appzoo yet might do that tonight.
    Thanks so much for you step by step guide PINTO would not have been able to rebuild without it.

Sign In or Register to comment.