Ayvazofsky
Posts: 10
Joined: Fri Aug 03, 2018 6:25 am

Troubles with YOLO on Raspberry Pi 3 B+

Fri Aug 03, 2018 6:51 am

Hi everyone recently I bought Raspberry Pi 3 B+ and install Raspbian I compile YOLO and try to run it, but when i run program i get Under-voltage detected! (0x00050005) and program doesn't run. Can anybody help me solve this problem? Who try YOLO on Raspberry? Any answer can help.
Thank you in advance.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27964
Joined: Sat Jul 30, 2011 7:41 pm

Re: Troubles with YOLO on Raspberry Pi 3 B+

Fri Aug 03, 2018 9:54 am

Use a better power supply to get rid of the undervoltage.

As for Yolo not running, if the Pi is dying due to low power that would stop it working, but I suspect it might be something else. Sort out the power supply first, then we can progress.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

Ayvazofsky
Posts: 10
Joined: Fri Aug 03, 2018 6:25 am

Re: Troubles with YOLO on Raspberry Pi 3 B+

Fri Aug 03, 2018 7:26 pm

Thanks a lot...I have corrected it...and received 0.1 FPS(( :o :oops:
Is'it maximum for Raspberry Pi?

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27964
Joined: Sat Jul 30, 2011 7:41 pm

Re: Troubles with YOLO on Raspberry Pi 3 B+

Fri Aug 03, 2018 9:31 pm

What is Yolo and what are you trying to do with it? If it is very CPU intensive, or memory hungry, then you have to remember the Pi is not a 8 core $1000 desktop with 16GB of RAM.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

Ayvazofsky
Posts: 10
Joined: Fri Aug 03, 2018 6:25 am

Re: Troubles with YOLO on Raspberry Pi 3 B+

Sun Aug 05, 2018 9:10 am

Is it impossible to obtain normal real time object recognition on raspberry? (at least 8 FPS). Thanks.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27964
Joined: Sat Jul 30, 2011 7:41 pm

Re: Troubles with YOLO on Raspberry Pi 3 B+

Sun Aug 05, 2018 10:10 pm

Well, you need to find out what sort of CPU instructions per second would be required, then compare that with the Pi performance figures. I have no idea about Yolo, so difficult to make a judgement.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

Ayvazofsky
Posts: 10
Joined: Fri Aug 03, 2018 6:25 am

Re: Troubles with YOLO on Raspberry Pi 3 B+

Mon Aug 06, 2018 5:31 am

Thanks a lot Jamesh...

MouseNJ
Posts: 1
Joined: Sun Aug 19, 2018 6:03 am

Re: Troubles with YOLO on Raspberry Pi 3 B+

Sun Aug 19, 2018 6:17 am

Ayvazofsky wrote:
Fri Aug 03, 2018 7:26 pm
Thanks a lot...I have corrected it...and received 0.1 FPS(( :o :oops:
Is'it maximum for Raspberry Pi?
Do you mean YOLO( You only look once), the convolution neural network algorithm to detect objects in a video stream? 0.1 FPS is very slow because of the limited RAM and computation speed. Image size and model parameter complexity also matters. I have little experience in optimization but you can have a try. Are you using MXNET?

kdupreez
Posts: 1
Joined: Tue Aug 21, 2018 6:05 am

Re: Troubles with YOLO on Raspberry Pi 3 B+

Tue Aug 21, 2018 6:07 am

Even with YOLO "Tiny" and a single class trained network configured for small 256x256 network input it takes 4.2 seconds per image..

VERY slow on RP3B+ without any GPU :(

Ayvazofsky
Posts: 10
Joined: Fri Aug 03, 2018 6:25 am

Re: Troubles with YOLO on Raspberry Pi 3 B+

Thu Aug 23, 2018 10:54 am

MouseNJ wrote:
Ayvazofsky wrote:
Fri Aug 03, 2018 7:26 pm
Thanks a lot...I have corrected it...and received 0.1 FPS(( :o :oops:
Is'it maximum for Raspberry Pi?
Do you mean YOLO( You only look once), the convolution neural network algorithm to detect objects in a video stream? 0.1 FPS is very slow because of the limited RAM and computation speed. Image size and model parameter complexity also matters. I have little experience in optimization but you can have a try. Are you using MXNET?
No I'm not used MXNET...but its look like I found solution. Using SSD (Single Shot Detection) and Movidius NCS Stick you can get 6-8 maybe 10 FPS...And i think that's great.

DanielChai
Posts: 15
Joined: Mon Jan 21, 2013 5:29 pm

Re: Troubles with YOLO on Raspberry Pi 3 B+

Wed Aug 29, 2018 3:13 am

First, don't think about running Yolo full version on RPI! Forget it!

Second, running Yolo tiny is totally possible. There are some optimization you can do to improve its FPS on RPI. The original darknet had poor performance on CPU (it was designed to run on CUDA GPU). You have to compile the darknet code with NNPACK to do parallel computation on multicore CPU.

Still, no way you can get 8-9 FPS without additional hardware like Intel Movidius.

I successfully ran Yolo v3 tiny on my Pi 3 model B+ and get 1FPS. Not too bad for trivial usage..
Check the results here
http://funofdiy.blogspot.com/2018/08/de ... real.html
Image

Ayvazofsky
Posts: 10
Joined: Fri Aug 03, 2018 6:25 am

Re: Troubles with YOLO on Raspberry Pi 3 B+

Wed Aug 29, 2018 11:59 am

DanielChai wrote: First, don't think about running Yolo full version on RPI! Forget it!

Second, running Yolo tiny is totally possible. There are some optimization you can do to improve its FPS on RPI. The original darknet had poor performance on CPU (it was designed to run on CUDA GPU). You have to compile the darknet code with NNPACK to do parallel computation on multicore CPU.

Still, no way you can get 8-9 FPS without additional hardware like Intel Movidius.

I successfully ran Yolo v3 tiny on my Pi 3 model B+ and get 1FPS. Not too bad for trivial usage..
Check the results here
http://funofdiy.blogspot.com/2018/08/de ... real.html
Image
Thanks I'll take your advice into account))
Try to get 8 FPS)

victagayun
Posts: 129
Joined: Sun Apr 15, 2012 4:01 pm
Contact: Website

Re: Troubles with YOLO on Raspberry Pi 3 B+

Thu Nov 15, 2018 6:47 am

I logged new issue on github.

https://github.com/shizukachan/darknet-nnpack/issues/18

@ Install qmkl

Code: Select all

sudo apt-get install flex
git clone https://github.com/Terminus-IMRC/qpu-assembler2
cd qpu-assembler2
cmake .

Code: Select all

CMake Error at CMakeLists.txt:22 (message):
  qasm2 not found.  Install it from
  https://github.com/Terminus-IMRC/qpu-assembler2


-- Configuring incomplete, errors occurred!
See also "/home/pi/qmkl/CMakeFiles/CMakeOutput.log".
Follow, clone and make https://github.com/Terminus-IMRC/qpu-assembler2
and install flex.

Code: Select all

sudo apt-get install flex
Then got this 2nd error

Code: Select all

pi@raspberrypi:~/qmkl $ cmake .
CMake Error at CMakeLists.txt:27 (message):
  qbin2hex not found.  Install it from
  https://github.com/Terminus-IMRC/qpu-bin-to-hex


-- Configuring incomplete, errors occurred!
See also "/home/pi/qmkl/CMakeFiles/CMakeOutput.log".
Follow and clone and make https://github.com/Terminus-IMRC/qpu-bin-to-hex

The last error I cannot solve anymore.

Code: Select all

pi@raspberrypi:~/qmkl $ cmake .
-- Checking for module 'libmailbox>=2.0.0'
--
CMake Error at /usr/share/cmake-3.7/Modules/FindPkgConfig.cmake:415 (message):
  A required package was not found
Call Stack (most recent call first):
  /usr/share/cmake-3.7/Modules/FindPkgConfig.cmake:588 (_pkg_check_modules_internal)
  CMakeLists.txt:52 (pkg_check_modules)


-- Configuring incomplete, errors occurred!
See also "/home/pi/qmkl/CMakeFiles/CMakeOutput.log".
I did this on fresh Raspbian.

Drallert
Posts: 1
Joined: Thu Nov 22, 2018 8:22 pm

Re: Troubles with YOLO on Raspberry Pi 3 B+

Thu Nov 22, 2018 8:29 pm

Ayvazofsky wrote: Is it impossible to obtain normal real time object recognition on raspberry? (at least 8 FPS). Thanks.
I've managed to get it to ~0.5 fps compiling with openmp option activaded, but that's still not enough
We'll have to figure out somehow, like this guy: https://www.youtube.com/watch?v=nTQHUsKE-xo
He doesn't give documentation though

DanielChai
Posts: 15
Joined: Mon Jan 21, 2013 5:29 pm

Re: Troubles with YOLO on Raspberry Pi 3 B+

Wed Dec 19, 2018 7:44 pm

If I have to guess, he might be using some USB deep learning accelerator like this
https://www.youtube.com/watch?v=MK7XExpfm7Q

Drallert wrote:
Thu Nov 22, 2018 8:29 pm
Ayvazofsky wrote: Is it impossible to obtain normal real time object recognition on raspberry? (at least 8 FPS). Thanks.
I've managed to get it to ~0.5 fps compiling with openmp option activaded, but that's still not enough
We'll have to figure out somehow, like this guy: https://www.youtube.com/watch?v=nTQHUsKE-xo
He doesn't give documentation though

Return to “C/C++”