Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

where can I found a proper guide/ tutorials or support for NCS

I was wondering where I can find a instruction / tutorial of how to create a simple model with tensorflow/keras for NCS or retrained a Tensorflow model for NCS? Got sick of tutorial with trained model.
I searched a lot, shockingly I could not find any guide which is working and really applicable.


  • 6 Comments sorted by Votes Date Added
  • edited November 2018 Vote Up1Vote Down

    You can't, don't even bother.
    You can not do anything with models trained on Tensorflow, and training on cafe is near impossible - 0 tutorials, 0 help, nothing matches with anything, an absolute disaster.

    Just sell the stick and run tensorflow models on your pc with GPU.

  • Try looking at open vino

  • edited December 2018 Vote Up0Vote Down

    Thank you for your reply
    I already found that open vino would be another way, but I dont feel confident and comfortable to use that, as I am not willing to rely on Intel stuff! (NCS has already killed lots of my time and that is enough to me)
    that is why I asked that there is any way to find a design/retrain a tensorflow model tutorial somewhere.

  • @KoitRK I feel exactly the same, I work on an AI project in a company for object detection, I was working on server-based GPU, unfortunately I met the NCS and start thinking to have something at the edge. After almost a month of working, I could not reach my goal to train my own model. and I met my project deadline and nothing in my hand.
    I also watched their tutorials, they even skip training model for NCS in their tutorials! :))))

  • I am totally with you on the traing step. I have also asked repeatedly for the same thing. As far as the Intel comment, you know they own movidius? For what the stick can do, i find it brilliant but immature tech. Good thing its inexpensive.
    A 1080 is 700$ takes a lot of watts and cant be put into a bot. And let's not forget Tensorflow has issues as well and don't get me started on jetson tx. What can anyone say the industry and tools are constantly changing. Most tools are less than 3 years old and still evolving. In short with so much fluidity i would not gamble deadlines unless you have done it before yourself since this is bleeding edge tech.

  • @chicagobob123 since I was thinking it should work as they promoted in their "examples", I bought it.
    yes you are right, the device is not that much expensive, but my time is valuable to me!

This discussion has been closed.