Converting Tensorflow Model: pb file to IR

[Rashik]

I am trying to create IR from frozen inference_graph.pb file for my tensorflow model. But it shows this error-

[ ERROR ] Exception occurred during running replacer "None" (<class
'extensions.front.tf.assign_elimination.AssignAddElimination'>): Data flow edge coming out of
AssignAdd node Adam

I am not sure how to solve it. Please help me

[Esteban_D_Intel]

Hello Rashik,

Thank you for posting on the Intel® communities.

To better assist you I would appreciate if you could provide the full model or part number of the Intel component that you are using for this design. 

We will be waiting for your response to find the appropriate support for your inquiry.

Esteban D.

Intel Technical Support Technician  

[Rashik]

I am using Intel Neural Compute Stick 2. Thank you.

[Esteban_D_Intel]

Hello Rashik,

Thank you so much for the information provided.

The thread will be routed to the appropriate support in order to find a resolution to this matter.

Esteban D.

Intel Technical Support Technician

[IntelSupport]

Hi Rashik,

Thank you for reaching out to us. If possible, could you share the frozen_inference_graph.pb file here? We would like to replicate this issue on our end.

Regards,

Adli

[Rashik]

Hello

I have attached the frozen inference_graph.pb file here. Thank you

Regards
Rashik

[Rashik]

Sorry I failed to attach the .pb file in the previous reply as .pb file is not supported. I tried to zip the file but it exceeds the maximum file size. if you can give me an email address, I can share the .pb file. Thank you.

[IntelSupport]

Hi Rashik,

I would suggest you share the file through cloud storage services such as dropbox and google drive. Then, copy and paste the link here.

Regards,

Adli

[Rashik]

https://drive.google.com/file/d/1hYBH8-qF6ykUqfM9jvgT_pwtNl24SoQ3/view?usp=sharing 

Here it is. Thank you.

[IntelSupport]

Hi Rashik,

Thank you for sharing the file. If possible, could you share additional information regarding:

• Input shape of the model?
• Which topology used for training? Is it an SSD model?

Regards,

Adli

[Rashik]

Hello

1. The input shape is 256*256
2. It is not an SSD model. I have used convolutional neutral network (convolution layer, pooling layer, residual unit, ReLU layer, fully connected layer) for image classification.

[IntelSupport]

Hi Rashik,

I have replicated the issue using OpenVINO 2020.4 version and OpenVINO 2021.1 version and I’m also seeing the same error in both versions.

So, the issue lies with your model. I would suggest you recheck the inference graph file. Make sure you follow the correct instructions on how to freeze a graph.

Having said that, step-by-step instructions on how to load a non-frozen model to the model optimizer is available at the following link: https://docs.openvinotoolkit.org/2021.1/openvino_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html#loading-nonfrozen-models

Besides, I am giving a link here which will be helpful for you to solve your issue: https://github.com/openvinotoolkit/openvino/issues/98 

Regards,

Adli

[Rashik]

Hello

I tried to follow all the instruction provided on how to freeze a graph. If you want, I can share my python code with you.

Another thing, do I need to use the checkpoint file to convert my model? If so, then can you please tell me how to create checkpoint_file.ckpt? Thank you.

Regards
Rashik Shadman

[IntelSupport]

Hi Rashik,

One of the ways to load a non-frozen graph to the model optimizer is the 'checkpoint' method. In this case, you need two files, an inference graph file (inference_graph.pb or inference_graph.pbtxt) and a checkpoint file. For more information, please refer to the following link:

https://docs.openvinotoolkit.org/2021.1/openvino_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html#loading-nonfrozen-models

In order to create the checkpoint file, please refer to the following link:

https://www.tensorflow.org/guide/checkpoint#writing_checkpoints

Regards,

Adli

[IntelSupport]

Hi Rashik,

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.

Regards,

Adli