Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.

Getting started without NCS?

idata
Employee
1,093 Views

Hello all,

 

So as the subject line suggests, I don't yet have the NCS as yet, hopefully within the next few weeks. Mean time I want to keep some sort of momentum and learn stuff. I started with the Software install following instructions at [1]. Unfortunately the second step began with "Plug the NCS to the host machine" :-( I didn't think I needed the NCS for the compile of a network but there you go.

 

So in the mean time I'm a newbie. So what to do? I did find blog post about setting up caffe, [2] but that was a dumb list of steps without explaining what you were doing. I did read an article about caffe which had a prerequiste of knowing Google's Protocol Buffers so I read up on that. (Not sure the world needs another protocol for seralising and defining data structures but hey ho.). So then I found a list of books [3]. I started at title 3 "Understanding Machine Learning: From Theory to Algorithms" but there was a bit of maths formula in it so I should go back and do the first in the list "Think Stats: Probability and Statistics for Programmers". Oh the joy.

 

No I need to find a good intro to caffe, defining a simple network to solve an XOR dataset. Define it, compile it, train it, execute it. What are the options for configuring a network, the algorithms for feeding forward and backward, if that's the right terminology. What happens to my XOR Net if I change from a petrol engine to Diesel to Electric. And what about the gear box, (transmission for ye in the US). What exactly are the engine and gear box options? And what do they mean in practice. I'm not a Car Designer but if somebody asked me to design a car I'd have half a clue, of the bits you need, I don't have that info for Machine Learning.

 

Has somebody found that information out there? If so lend a hand here and point me to it. Sometimes putting together a good search term for the Internet is all you need and I've failed.

 

[1] https://developer.movidius.com/getting-started/compile-networks-using-the-toolkit

 

[2] https://www.oreilly.com/learning/how-to-build-and-run-your-first-deep-learning-network

 

[3] http://www.kdnuggets.com/2017/04/10-free-must-read-books-machine-learning-data-science.html
0 Kudos
3 Replies
idata
Employee
710 Views

@jwhitmore if you followed steps 1 thru 3 from https://developer.movidius.com/getting-started, you have successfully installed MvNC_SDK on your system. Here's how the SDK is organized:

 

MvNC_Toolkit

 

     

  • Compiler (Does not need physical HW) - Translation of network weights and structures from Caffe deep learning frameworks into a NCS compatible format
  •  

  • Profiler (Needs physical HW) - The generation of detailed, per-layer performance statistics of how your network is running on the NCS.
  •  

  • Checker (Needs physical HW) - Verification of classification accuracy by running inferences with the NCS.
  •  

 

MvNC_API (Needs physical HW) - Used to write applications in Python or C/C++ that can load graph file on the NCS, and then trigger inference on the NCS.

 

https://developer.movidius.com/getting-started is a basic set of instructions you need to ensure your MvNC_SDK installation has succeeded, and that your host can offload inference to the NCS. Please visit the software download section for a detailed explanation of how to use toolkit and API. Ex:

 

Toolkit doc - https://ncs-forum-uploads.s3.amazonaws.com/ncsdk/MvNC_SDK_01_07_07/NCS_Toolkit_1.07.06.pdf

 

API doc - https://ncs-forum-uploads.s3.amazonaws.com/ncsdk/MvNC_SDK_01_07_07/NCS_API_1.07.07.pdf

 

As for generic deep learning learning material, I find this youtube channel very useful for beginners - www.youtube.com/channel/UC9OeZkIwhzfv-_Cb7fCikLQ

 

Once you have gone through the first couple episode on this channel, the oreilly book you pointed out seems a lot more easier to follow!

0 Kudos
idata
Employee
710 Views

Thanks a million for that help. I've been searching around and found a book which looks great at introducing the concepts but the problem is in the title: http://shop.oreilly.com/product/0636920052289.do (Tenserflow). I think http://karpathy.github.io/neuralnets/ is a great web page.

 

Thanks for the pointer to the docs I want to start compiling things, even if my NCS hasn't arrived as yet, if I do too much reading it'll all get too theoretical. There's a balance I guess.

0 Kudos
idata
Employee
710 Views

 

if I do too much reading it'll all get too theoretical.

 

I hear you @jwhitmore, back in college I enjoyed my practical exams more than my written tests :# . Btw thanks for sharing those links!

 

0 Kudos
Reply