OpenFace in a Raspberry Pi 3 (ARM)

[Gorka]

Hi,

First of all thank you guys for this incredible open source project!

I've been using it on my Macbook and it works really well. Now I have a project where I need to make it run on a Raspberry Pi 3 with an ARMv8 processor. I already got Docker setup but I was not able, although I tried in many different ways, to make it run with the "bamos/openface" image. The SO is Raspbian Jessie (based on Debian Jessie) and the Docker version is 1.10.3.

Could you please help making this work on the Pi?

Thank you a lot for your time!

Regards,

Gorka

[Gorka]

I also tried building it from the repository as described in the Setup Guide, but I got an "exec format error", the same I get if I use the "bamos/openface" image and try to run it.

P.S.: I mistyped Operating System. I wrote SO where I wanted to write OS.

Thank you!

Gorka

[Brandon Amos]

Hi Gorka, since I built the Docker container on x86, it contains

some x86-specific code that won't execute on ARM.

I recommend:

  1. Try to build the Docker container from scratch, which will require

  re-building the base opencv-dlib-torch image. If you don't change the base

  image you're still pulling in compiled x86 code.

 

  First build:

    https://github.com/cmusatyalab/openface/blob/master/opencv-dlib-torch.Dockerfile

  Then change the FROM to that container and build:

    https://github.com/cmusatyalab/openface/blob/master/Dockerfile

  2. Build the dependencies from scratch for ARM, not using Docker.

-Brandon.

[Gorka]

Hi Brandon,

Thank you a lot for your help! I will try it in a few hours.

Thank you!

Gorka

[Gorka]

Hi Brandon,

I changed the "opencv-dlib-torch.Dockerfile" to be built from the "resin/rpi-raspbian" image and it was all ok for like 3 hours until it started compiling OpenCV and it threw many errors related to H264, MPEG-2, etc. Do you know what may be causing it?

Thank you a lot!

Regards,

Gorka

[basni...@gmail.com]

Hi all,

I am also trying to get OpenFace to work on a Raspberry PI 3.

Did you succeed? If yes, could you please share how you did it and how the performance is.

Best, Bas

[YC]

Have you installed the related packages? For example:
sudo apt-get -qq install libopencv-dev build-essential checkinstall cmake pkg-config yasm libjpeg-dev libjasper-dev libavcodec-dev libavformat-dev libswscale-dev libdc1394-22-dev libxine-dev libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev libv4l-dev python-dev python-numpy libtbb-dev libqt4-dev libgtk2.0-dev libfaac-dev libmp3lame-dev libopencore-amrnb-dev libopencore-amrwb-dev libtheora-dev libvorbis-dev libxvidcore-dev x264 v4l-utils ffmpeg cmake qt5-default checkinstall

[nitish bhardwaj]

Hi,

Has anyone successfully installed Torch on PI ?

Please check the issue at https://github.com/torch/nn/issues/803 

[Vijayenthiran]

Hi,

Did you tried installing directly without using the docker? I was able to successfully setup and run open face in odroid-xu4.

[borch...@gmail.com]

Hello,

someone was finally able to docker built for Raspberry 3?


Thanks
.axel

[William Chen]

Hi,

I compiled the Dlib, which was used for detection and landmark generation for open face, on the Pi 3 B. Immediately, I noticed a huge performance drop compared to the version I compiled on x86. Processing a 500 by 375 image will take approximately 16 s on the Pi ARM chip (1.2 GHz) compared to ~ 300 ms on the intel x86 chip (2.8 GHz). Then I compare Dlib on two CPUs has similar clock speed (x86 vs ARM) at about 540 MHz. x86 finished it in 400 ms where as ARM took about 90 s on a same resolution image.

I've read somewhere that we need to enable SIMD instructions on the ARMv8 chip. Just wondering if there is anyone has done that before.

[joov]

Docker Image is joov/rpi-openface

[fatih şerefli]

Hello Vijayenthiran
i am trying instal openc face on Odroid ux4 can you help me about this subject ?
i would be glade


28 Temmuz 2016 Perşembe 09:01:26 UTC+3 tarihinde Vijayenthiran yazdı:

[Estérfano Emanoel Marques Lopes]

Has someone get results? I have trying for a week extract the landmarks points from a face, but dlib give me a ARM error, more specific something about 64 bits.

I'm trying on a Raspberry Pi 3 Model B. Someone get success on this try?

[Igor Moiseev]

It is a bit late, but currently it was solved within dlib library. Here is the key comment https://github.com/davisking/dlib/issues/557#issuecomment-298230416