Error convert TF2 model

[Elul__Eliav]

Hi,

I try to convert Mobilenet model was retrained with TF2 to OpenVino model.

I use the "SavedModel" loading for loading the TF model with this command:

"python mo.py --framework tf --saved_model_dir <Path to model>\model --model_name ir_1 --output_dir <Path to model>\Out"

and I get this error:

[ ERROR ]  Unexpected exception happened during extracting attributes for node conv_pw_13_bn/moving_variance/Read/ReadVariableOp.
Original exception message: 'ascii' codec can't decode byte 0xca in position 1: ordinal not in range(128)

When I converted similar model with frozen graph loading and Tensorflow 1.13 everything was right.

Is known about a problem with Mobilenet layers that was trained with TF2?

- Tested with OpenVino 2019 R3.1 and 2019 R2

Thanks,

Eliav

[JesusE_Intel · Accepted solution]

Hi Eliav,

Thanks for reaching out. Currently the Model Optimizer does not support Tensorflow 2.0. However, there has been reports by other users that if you use Tensorflow 1.14 freeze_graph.py to freeze a Tensorflow 2.0 model the Model Optimizer may accept it.

Hope this helps.

Regards,

Jesus

[Anonymous]

This is not true at all! I have wasted another day based on the fact Intel staff have said that you can convert TF2 models by freezing them in TF1.14. It does not work.