Recommended way to export models in TensorFlow 2.0 for using along with OpenVINO model optimizer

[SPaul19]

Hello. I am currently porting the codebase of one of my projects to TensorFlow 2.0. In the past, I have had experience in deploying fine-tuned Keras models on NCS using OpenVINO (with TensorFlow 1.14). Is there any recommended way to do that when it comes to TensorFlow 2.0 since the way we export TensorFlow-formatted models (typically protobuffers) is changed in v2. So, directions regarding this would be great. 

[Shubha_R_Intel · Accepted solution]

Dear Sayak

Currently Model Optimizer does not support Tensorflow 2.0. But if you use Tensorflow 1.14 freeze_graph.py to freeze a TF 2.0 model, this frozen pb should be accepted by Model Optimizer.

Hope it helps,

Thanks,

Shubha

[Maroquio_Bernardo__R]

Hello, Subha.

Do you have any news about convert TensorFlow 2.0 to OpenVINO / NCS2?

[ezeng3]

Hello, Subha:

            Is there any time table from Intel that openVino will support tensorflow 2.x?

Best Regards

Edwin Z

[JesusE_Intel]

Hi Edwin/Ricardo,

Unfortunately, I am unable to comment on this as Intel does not disclose roadmaps. I encourage you to check the Release Notes when there is a new software release. 

Regards,
Jesus