Caffe for embedded devices

1,515 views
Skip to first unread message

ad...@buddyguard.io

unread,
Jul 10, 2015, 8:41:08 AM7/10/15
to caffe...@googlegroups.com
Hei there,

I'm wondering if anyone tried to integrate Caffe on a embedded device with limited resources, running Linux. The task for Caffe would be to do face recognition and few object detections (2 different networks). I observed that there were some adaptions for mobile devices, such as Android or iOS, but no adaptation for a low-powered embedded device.

Could anyone tell me if Caffe is viable to be run on one of the following platform specifications, or I should go with other frameworks/libraries, such OpenCV, etc.
 

1 - Single core processor of 1Ghz (i.MX 6Solo  from FreeScale),  512 Mb Ram


2 - Dual Core processor of 1,2Ghz (i.MX 6DualPlus from FreeScale), 1Gb Ram


3 - Quad Core processor of 1,2Ghz (i.MX 6Quad  from FreeScale),  1Gb Ram


It would be also interesting to mention, for which task would be viable (face recognition or object detection).


I will get next week a Nitrogen platform, and I'll try to do some testing over there - http://boundarydevices.com/product/nitrogen6x-board-imx6-arm-cortex-a9-sbc/


Looking forward for opinions & suggestions!



 

 


Corey Thompson

unread,
Jul 10, 2015, 1:43:40 PM7/10/15
to caffe...@googlegroups.com
I suggest the NVIDIA Jetson board.  It supports CUDA (which is really well supported and will save you a lot of time), and would likely meet all of your embedded power needs.  See https://developer.nvidia.com/jetson-tk1  and http://elinux.org/Jetson_TK1.

Going with caffe is a great idea, and works on the Jetson:

best!
Corey

ad...@buddyguard.io

unread,
Jul 13, 2015, 6:51:22 AM7/13/15
to caffe...@googlegroups.com
Corey, thanks for the suggestion! Unfortunately, my requirement is to run Caffe on a new embedded device (which we build), and which is gonna have the resources mentioned on the main post. Thus, Nvidia Jetson Board (which is a really powerful one) is out of discussion for now :-(

That's why I was interested in the hardware specifications for embedded devices. 

Cheers,
George

Jonathan Laserson

unread,
Aug 11, 2015, 5:45:31 PM8/11/15
to Caffe Users
George, I'm facing the same challenge, and I was wondering what were your finding regarding using caffe on ARM.

              Jonathan

林子達

unread,
Nov 11, 2015, 9:50:42 AM11/11/15
to Caffe Users
Hi 
I implemented Caffe and fastrcnn on Android platform. You can try it on arm device

林子達

unread,
Nov 11, 2015, 9:50:58 AM11/11/15
to Caffe Users

https://github.com/tzutalin/Android-Object-Detection
林子達於 2015年11月11日星期三 UTC+8下午10時50分42秒寫道：

Ninja At Work

unread,
Nov 11, 2015, 1:21:23 PM11/11/15
to Caffe Users
George, I'm facing the same challenge, and I was wondering what were your finding regarding using caffe on ARM.
Saeed :)

Lee T

unread,
Oct 9, 2016, 3:52:05 AM10/9/16
to Caffe Users
Is there anyone have solved this problem? I am really want get some helps. 
The resource of my embedded device is so limited:
Single core processor of 786 MHz, 128 M  RAM, embedded Linux operation system without compiler (gcc, g++).
I want to cross-compile caffe using cross tool chain, but I am not familiar with cross-compile ......
:-(
help, help

在 2015年7月10日星期五 UTC+8下午8:41:08，ad...@buddyguard.io写道：
Reply all
Reply to author
Forward
0 new messages