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.

Convert a custom model to IR.

dilip96
New Contributor I
1,049 Views

Hi,

 

This post is to gather some information regarding the IR formats.

 

I have a model built using the TensorFlow. This model takes the inference results ( label Id and B Box coordinates ) from a Object Detection Model and provides result based on some simple calculations. 

So , This Model does not have any conventional Deep Learning layers. All it does is perform some kind of addition multiplication and slicing operations on the input and store it as different tensors. Also the Model takes input in the form of python list.

Now, I am trying to figure out if this Model (without any CNN layers) can be converted to IR formats? or what are the requirements for IR formats so that I can convert model like this?

 

I was also wondering if it is possible to develop this kind of model in OpenVINO itself using nGraph? If yes which method will you advice?

 

I am going through the OpenVINO docs but it takes a lot of time to understand them. So if you can provide me with some quick answers and then direct me towards proper documents it will be great.

 

Please feel free to reply back if I have not made my question clear or if its too confusing.

 

Thank you.

0 Kudos
5 Replies
Zulkifli_Intel
Moderator
1,017 Views

Hello Dilip Ravi,


Thank you for reaching out to us.

 

We are checking into this and get back to you with the information regarding this issue soon.

 

Sincerely,

Zulkifli 


0 Kudos
Zulkifli_Intel
Moderator
931 Views

Hello Dilip Ravi,


Sorry for the delay in response. Based on Model Optimization Techniques documentation. CNN layer is required when you want to optimize the model unless you choose to turn off the optimization function by adding this --disable_fusing when converting the model to IR format.

 

Sincerely,

Zulkifli


0 Kudos
dilip96
New Contributor I
914 Views

Thank you Zulkifli for your response.

The reason I asked this question is, 

I am trying to develop a centroid tracking algorithm as an Model that can Run on Intel NCS2. Since the algorithm is pretty much basic calculations along with slicing and concating of Tensors, I wanted to be sure that this is possible in the first place. I am using TensorFlow framework for the same.

Please let me know your thoughts and inputs. Is this possible?

I appreciate your help.

0 Kudos
Zulkifli_Intel
Moderator
802 Views

Hello Dilip Ravi,


The conversion of the algorithms you mentioned is not validated yet, but as all of them seems to be implemented using neural network, you could try converting them using general guidelines. Please note that this is not guaranteed it will work.

 

Here is the Overview of OpenVINO™ Toolkit Intel's Pre-Trained Models that you can explore and get some ideas on how to implement them.

 

Sincerely,

Zulkifli 


0 Kudos
Zulkifli_Intel
Moderator
771 Views

Hello Dilip Ravi,


This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question.


Sincerely,

Zulkifli


0 Kudos
Reply