USE ONNX Problem

[CLi37]

I have ported a Tensorflow 2 model to work on DevCloud.

However my model can not be saved as a complete file, only the weight file can be saved.

Then how can I save it to an ONNX package for prediction and distribution? 

[VaradJ_Intel]

Hi,

Thanks for posting in Intel communities.

Please can you let us know which DevCloud are you using?

Thank you.

[CLi37]

I use Intel JupyterLab with Intel Python and Tensorflow+ONNX kernel to train the model. I enabled ONNX with two-line codes.

I want to pack Python + Tensorflow trained model into the ONNX package and distribute on other ONNX devices? Any examples of doing so? 

[VaradJ_Intel]

Hi,

Please can you share the Intel DevCloud that you are using? Is it Intel DevCloud for oneAPI or FPGA or Edge?

Thanks

[CLi37]

It is Intel DevCould for the Edge. It is kernel is Python 3 (Tensorflow_OpenVINO 2022.1) in JuypterLab.

[Hari_B_Intel]

Hi CLi37

we might need some time to further investigate this issue. please allow us some time to investigate it.

Thank you

[Hari_B_Intel · Accepted solution]

Hi CLi37,

Thank for your patience and understanding, we have two examples of OpenVINO EP for ONNX Runtime.

Jupyter notebook -> Reference_samples -> iot-devcloud -> openvino-lts -> developer-samples -> python -> ONNX-Object-detection-python

and

Jupyter notebook -> Reference_samples -> iot-devcloud -> openvino-lts -> developer-samples -> onnx_openvino_classification

Both can be used with Python3 (OpenVINO 2021.4.2) Kernel

For your information, DevCloud is not meant for training and I don't think there is possible to generate the model.

You can still upload the complete save model to DevCloud for prediction and distribution.

Hope this information helps

Thank you