TensorFlow Object Counting API with OpenVINO

[n30]

Hi,

is there a good guide or tutorial on how to use the TensorFlow Object Counting API with OpenVINO, ideally on Raspberry Pi + the Intel Neural Compute Stick and ideally for custom objects using a frozen model in form of a .pb file.

I really tried to find something, but encountered only solutions for parts of it, which then do not work together. If anyone has any links, please let me know.

Many thanks o_O

[David_C_Intel]

Hi n30,

Thanks for reaching out. We do not have an official guide or tutorial for that specific API, but you can refer to this Converting TensorFlow* Object Detection API Models documentation to try it and get an idea on how it works. Also, there might be other projects such as this People Counter that use OpenVINO™ toolkit and Intel® NCS2, which you can check and modify to suit your needs.

Best regards,

David C.

[n30]

Dear David,

thank you for your reply.

I followed the Udacity tutorial. 

However, when I try to convert my .pb file to the IR format using the model optimizer, I get the following error:

[ FRAMEWORK ERROR ] Cannot load input model: TensorFlow cannot read the model file: "retractedpath/model_optimizer/model.pb" is incorrect TensorFlow model file.
The file should contain one of the following TensorFlow graphs:
1. frozen graph in text or binary format
2. inference graph for freezing with checkpoint (--input_checkpoint) in text or binary format
3. meta graph

Make sure that --input_model_is_text is provided for a model in text format. By default, a model is interpreted in binary format. Framework error details: Wrong wire type in tag..
For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #43.

Is it possible this is because I exported the .pb from Google Cloud AutoML Vision API.

I used the "Export your model as a TF Saved Model to run on a Docker container." export function. Is there any extra step here - it is a normal .pb file. There is no export for plain TensorFlow, only this and TensorFlowJS and TensorFlow Lite.

Thanks!

[n30]

Dear David,

thank you for your reply.

I followed the Udacity tutorial. 

However, when I try to convert my .pb file to the IR format using the model optimizer, I get the following error:

[ FRAMEWORK ERROR ] Cannot load input model: TensorFlow cannot read the model file: "retractedpath/model_optimizer/model.pb" is incorrect TensorFlow model file.
The file should contain one of the following TensorFlow graphs:
1. frozen graph in text or binary format
2. inference graph for freezing with checkpoint (--input_checkpoint) in text or binary format
3. meta graph

Make sure that --input_model_is_text is provided for a model in text format. By default, a model is interpreted in binary format. Framework error details: Wrong wire type in tag..
For more information please refer to Model Optimizer FAQ question #43.

Is it possible this is because I exported the .pb from Google Cloud AutoML Vision API.

I used the "Export your model as a TF Saved Model to run on a Docker container." export function. Is there any extra step here - it is a normal .pb file. There is no export for plain TensorFlow, only this and TensorFlowJS and TensorFlow Lite.

Thanks!

[n30]

Hi David,

thank you for your reply.

I followed the Udacity tutorial. 

However, when I try to convert my .pb file to the IR format using the model optimizer, I get the following error:

[ FRAMEWORK ERROR ] Cannot load input model: TensorFlow cannot read the model file: "retractedpath/model_optimizer/model.pb" is incorrect TensorFlow model file.
The file should contain one of the following TensorFlow graphs:
1. frozen graph in text or binary format
2. inference graph for freezing with checkpoint (--input_checkpoint) in text or binary format
3. meta graph

Make sure that --input_model_is_text is provided for a model in text format. By default, a model is interpreted in binary format. Framework error details: Wrong wire type in tag..

Is it possible this is because I exported the .pb from Google Cloud AutoML Vision API.

I used the "Export your model as a TF Saved Model to run on a Docker container." export function. Is there any extra step here - it is a normal .pb file. There is no export for plain TensorFlow, only this and TensorFlowJS and TensorFlow Lite.

Best

[David_C_Intel]

Hi n30,

Could you please answer the following:

• From which topology is your model?
• Could you share your .pb file and the command you used to convert it to IR format, so we can test it from our end?

Regards,

David C.

[David_C_Intel]

Hi n30,

If you have any additional questions, please submit a new thread as this discussion will no longer be monitored.

Best regards,

David C.