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.
6392 Discussions

How to compile TensorFlow CNN trained models with MVCN compile?

idata
Employee
766 Views

I followed this tutorial for real-time object detection using Movidius NCS. It worked perfectly.

 

However, I noticed that it used a Caffemodel in the tutorial. I have a trained TensorFlow model that I want to compile onto the Movidius. I would you be very grateful if you could point me to a tutorial on how I can achieve that.

 

Here is an example of model I want to compile.

 

Kind regards,
0 Kudos
2 Replies
idata
Employee
359 Views

You are asking something I have been asking for a while. HOW do you successfully generate a TensorFlow with your own custom classifications?

0 Kudos
idata
Employee
359 Views

@Samuel @chicagobob123

 

You need to use the mvNCCompile command in from the SDK. Not all models will be supported due to the limited ops support on the Movidius Neural Compute Stick.

 

There is a guide in the official documentation https://movidius.github.io/ncsdk/tf_compile_guidance.html

 

I have also written a brief guide for a personal project that is specifically what you're asking for. https://github.com/andrewginns/MSc-Project/blob/master/Movidius_Instructions.md

 

Basically you need to

 

1) Convert the .ckpt files to a frozen GraphDef

 

2) Convert that GraphDef into an inference optimised model

 

3) Run the mvNCCompile command to convert it to a movidius graph
0 Kudos
Reply