How to pre-train a model to be used with my own live data?

[Warren__Matt]

Hello,

Please pardon the triviality of my question, as I am new to Computer Vision and AI.

I was wondering how to use my existing database of plant photos to create a pre-trained model that I can use against live input (plant photos), to get the inference/ouput?

Any help/tips would be greatly appreciated!

Thank you!

Matt

[whatthisismyname]

TIPS:
1- Wrong Forum
2-You need to find a framework and suites you
3-Learn how to use it
4-Make sure its compatible with NCS

5-Train your model
6-Convert it to OpenVINO format 
7-Build your code  and be happy :)

[Warren__Matt]

Hello Daniel,

Thank you for the tips. I am hoping you can enlighten me:

1. If this is the wrong forum for learning how to use OpenVino, can you point me to the correct forum?

2. I am not sure what "You need to find a framework and suites you" means? I have built a Python script utilizing Tensorflow for my CNN algorithm(s) and I am getting decent accuracy - how do I translate that to something that can be consumed by the OpenVino Model Optimizer?

Thank you for your time!

Matt

[nikos1 · Accepted solution]

Hello Matt,

After you have a Tensorflow model the next step would be to generate valid IR (bin and xml files) that OpenVino can use

Please refer to https://software.intel.com/en-us/articles/OpenVINO-ModelOptimizer on more details how to use the model optimizer.

Inference Engine enables deploying your network model trained with any of supported deep learning frameworks: Caffe*, TensorFlow*, MXNet*, Kaldi*, or converted to the ONNX* format. To perform the inference, the Inference Engine does not operate with the original model, but with its Intermediate Representation (IR), which is optimized for execution on end-point target devices. To generate an IR for your trained model, the Model Optimizer tool is used.

A good start with OpenVino is also to build and run the samples as they demonstrate the whole workflow.

Cheers,

Nikos

.

[Warren__Matt]

Hello Nikos,

Thank you so much for this information. This bridges a major gap for me. I have my script, I have my data for training, I just don't know how to get this stuff into a (*.pb?) format for optimizing - the info you shared is a great start for me. Thank you.

As for starting with OpenVino, I have completed the 2 provided demos, but maybe I treated them as a black box. I will dig further into the contents.

Thank you again.

Matt

[nikos1]

Hi Matt,

I typically freeze Tensoflow models (please search how to freeze to pb) but it seems OpenVino can also work with not frozen models as shown in section "Load Non-Frozen Models to the Model Optimizer"  of  

https://software.intel.com/en-us/articles/OpenVINO-Using-TensorFlow

nikos

[Gdeep]

Dear Warren 

Please Follow this tutorial to train your tensorflow model on the custom object detection

https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/training.html

[Warren__Matt]

Hello Nikos and Gagan,

Firstly, thank you very much for taking the time to help me.

Secondly, I am going to proceed with these beginning steps and follow up.

Thank you very much, again.

-Matt

[whatthisismyname]

I guess you got your answer but you were on step #5 so yes, just converted the data and use it with openVINO sdk :)