[Dprglist] Pre-Order now: Google AIY Vision - Intel Movidius based kit

Thu Nov 30 19:10:04 PST 2017

 I did get the Movidius compute stick 'hello world' working but it's not exciting. I am trying to get yolo working on it with the raspberry pi.
thanks,
Alyssa
    On Thursday, November 30, 2017, 8:46:15 PM CST, Ezra Christensen <ezracc at gmail.com> wrote:  

 Have you tinkered with the Movidius compute stick? 
I ordered one and will let you know if I get a 'hello world' working for it.

- Ezra
------ Original Message ------From: "Carl Ott" <carl.ott.jr at gmail.com>To: "Alyssa Pipe" <eh.lyssa at yahoo.com>Cc: "DPRG D-List" <dprglist at dprg.org>Sent: 11/30/2017 7:10:00 PMSubject: Re: [Dprglist] Pre-Order now: Google AIY Vision - Intel Movidius based kit

I'm hoping the Google AIY Vision will be easier to work with than the Movidius USB compute stick.  After all, the Vision kit comes with instructions to assemble a cardboard box.  I bet the USB stick didn't have that much clarity and assistance...  
Sure - call me lazy, but I've been waiting for somebody else to put a 'plug-n-play' kit together for the Movidius.  I figure that if somebody else has worked out 'hello world', I can start kit-bashing and take it from there...
- Carl
ps. Where is this DIY stuff heading anyhow - maybe BioHacking kits from IKEA? Just follow the cartoon stick figures and make your own custom self-luminescent organic lighting decor?

On Thu, Nov 30, 2017 at 6:58 PM, Alyssa Pipe <eh.lyssa at yahoo.com> wrote:

 There is not much to the google AIY. It comes with a cardboard box, an arcade button, a speaker, a microphone board and an audio slash I/O hat for a raspberry pi. It does not include the raspberry pi, but I have half a dozen of those. The AIY is meant to make it easy for people to play with google's cloud based voice AI that the google home uses. I already 3d printed a replacement box for it, rather than the cardboard one it comes with and I have it setup currently to act just like a google home. 
The movidius stick is not quite as easy to use or setup. Their libraries are a little fussy to setup and use. I got it working on an ubuntu VM, and I want to use it on the raspberry pi but I don't have that part going yet.
The idea of the movidius stick is that it is a low power CPU optimized for running neural networks. Now anything can run a neural network but some architectures are better than others at that task because it involves lots of linear algebra. The movidius is relative fast for that task given it uses under 0.5 watters.
Thanks,
Alyssa
    On Thursday, November 30, 2017, 6:48:48 PM CST, Carl Ott <carl.ott.jr at gmail.com> wrote:  

 Although I actually haven't bothered to read the exact details of what the Google Kit does...  
as I understand it, this Movidius VPU has silicon neural nets optimized for full rate / full frame vision processing, with power requirements suitable for small battery powered things...
At the CES demo, they gave the impression that the host Pi would 'do little more than' set up the VPU, serve a video stream, and harvest results...
For our SmartCamBot, we could have used one of these to process the image locally, instead of having to rely on a network connected neural net running on IBM Watson. Except that, for that hackathon, using Watson was one of the prized criteria...
For DPRG, it seems likely this could revolutionize the challenge of spotting an orange can...
Just saying...  
I'm out to try new approaches...
Carl

On Thu, Nov 30, 2017 at 6:38 PM, Doug Paradis <paradug at gmail.com> wrote:

Carl,    Does this unit use the Zero's WiFi to send data to your laptop? What exactly does "local image processing" mean?
Regards,Doug P.
On Thu, Nov 30, 2017 at 6:23 PM, Carl Ott <carl.ott.jr at gmail.com> wrote:

I've been waiting for an eval kit like this since CES in January, when Dave Ackley and I saw a cool demo using the neural net based Movidius VPU - albeit that one was packaged as a USB compute stick.  This version runs on a bonnet for Raspberry Pi Zero W:

The VisionBonnet circuit board has an Intel Movidius MA2450 low-power vision processing unit, which can run neural network models right on the device. You'll get software, too, which has three TensorFlow-based neural network models: one to recognize a thousand common objects, another that can recognize faces and expressions and a third that can detect people, cats and dogs. 

Check this out:
https://www.movidius.com/solut ions/vision-processing-unithttps://uploads.movidius.com/1 463156689-2016-04-29_VPU_Produ ctBrief.pdf

Pre-Order from Micro-Center
https://www.engadget.com/2017/ 11/30/google-diy-ai-camera-kit -raspberry-pi/http://www.microcenter.com/sit e/content/Google_AIY.aspx?ekw= aiy&rd=1

Oh yeah. 
Google:
https://aiyprojects.withgoogle .com/vision 

I suppose, if we wanted to bring the award winning / beer finding SmartCamBot back to life with one of these, we'd have to train it to recognize bottles versus crayons, but that's a small price to pay for real-time / local image processing at this level...
Who else is in?
- Carl

______________________________ _________________
DPRGlist mailing list
DPRGlist at lists.dprg.org
http://lists.dprg.org/listinfo .cgi/dprglist-dprg.org

______________________________ _________________
DPRGlist mailing list
DPRGlist at lists.dprg.org
http://lists.dprg.org/ listinfo.cgi/dprglist-dprg.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dprg.org/pipermail/dprglist-dprg.org/attachments/20171201/fadb7f86/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Untitled
Type: image/png
Size: 887 bytes
Desc: not available
URL: <http://lists.dprg.org/pipermail/dprglist-dprg.org/attachments/20171201/fadb7f86/attachment.png>