Unable to optimize model to INT8 precision using DevCloud DL Workbench

[Afiq]

Hi, we did retraining using training extension https://github.com/openvinotoolkit/training_extensions/tree/develop/models/object_detection/model_templates/face-detection and then we export it, we convert it into precision FP32 and FP16. Also, we want to convert it into INT8 using DL Workbench on the DevCloud, but while i tried to optimize it into INT8 we face a problem saying that "Dev Cloud remote has failed". The retrained model that we used was face-detection-0200. How do i solve this problem?

[ChrisB_Intel]

Afiq,

Would you provide me the dataset, the model, and the method used so that we may reproduce the issue? Feel free to send me anything in a private message that you do not wish to share publicly.  

Thanks

Chris

[Afiq]

I have send you a private message regarding my problems and also i have attach the required file

[artyom_tugaryov]

Hello @Afiq 

Thank you for your question!

To understand what is the problem, can you share the DL workbench log file? 

How to download the log file you can find by the following link: https://docs.openvinotoolkit.org/latest/workbench_docs_Workbench_DG_Troubleshooting.html#download-logs

[Afiq]

here is the log

[artyom_tugaryov]

I reproduced this error in DL Workbench and using  POT  directly. This is a bug and we are fixing it. I hope the fix will be available on the next releases of DL Workbench (and OpenVINO of course)

[Afiq]

@artyom_tugaryov how do i know if the bug has already been fixed?

[JesusE_Intel]

Hi Afiq,

Is it possible for you to share your model in ONNX format for us to further test?

Regards,

Jesus