Still trying to understad how this device fits in

[idata]

First you cant even buy this device. I tried yesterday and am on the waiting list.

Next is it possible to get better than YOLO speeds on a low powered device? That is the desired result isn't it? Or am I wrong. I am looking for autonomous driving and need to mark locations of the standardaffair and some vehicles that are not on a common roadway.

[idata]

Hi @chicagobob123, I'm glad you got on the waiting list for the NCS. It sold fast at launch but I'm sure you'll be able to get yours soon.

Some users have gotten YOLO Tiny running on the NCS but its not something we validate with. One thing to keep in mind about the NCS is that its designed to stand out as a solution for low powered applications rather than strictly performance critical applications.

[idata]

Hi Neal,

I will be the guy asking all the dumb questions as I am rushing trying to understand TensorFlow, DNN and now this device. I will look for code that runs YOLO tiny. I will have to setup a machine to run Linux as that's not our core here, we are a Windows shop in the gold rush looking for DNN solutions. I am looking at NCS as a plugin assistant so that our embeded quad cores can actually RUN DNN. Getting one image analysis every second or so is not acceptable.Watched as many videos yesterday as I could. Whats the difference between the "fathom" and this device.The "fathom" had Tensor Flow support which runs under windows etc..

[idata]

Hi @chicagobob123, this device (Movidius NCS) is the follow on to the Fathom device. The Fathom was only released as a beta in limited quantities and this device is, of course, publicly available. Part of the Fathom beta included finding the most efficient path to production for the device. This process drove the initial feature list of the NCS which as you noted includes Caffe but not Tensorflow.

Neal