Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.
6061 Discussions

Not able to use customised model to create project and do INT8 calibration

Piyush_Khandelwal
223 Views

Hi Community,

 

My aim is to convert a custom model to INT8 IR v11 model using workbench but I'm unable to do so.

 

I am able to import the onnx model and since I want normalization I have defined Inputs section as well with means and scales. After doing these steps there is not option to go forward and create/evaluate the project and further to do INT8 calibration.

 

On the other hand if I don't specify Inputs section and avoid normalization then I'm able to use that model and create a project for INT8 calibration.

 

Attaching the screenshots for the same as well.

 

 

 

0 Kudos
2 Replies
Hairul_Intel
Moderator
189 Views

Hi Piyush_Khandelwal,

Thank you for reaching out to us.

 

I've validated from my end using squeezenet1.1 and mobilenetv2 from ONNX Model Zoo (This is an external link and is not maintained by Intel) and was able to perform INT8 calibration.

 

Here are the results of my findings:

 

Squeezenet1.1 ONNX model:

Screenshot from 2022-05-25 15-43-43.png

 

Screenshot from 2022-05-25 15-48-24.png

 

Mobilenetv2 ONNX model:

Screenshot from 2022-05-25 16-09-02.png

 

Screenshot from 2022-05-25 16-15-32.png

 

Additionally, please share your custom model with us and also provide more details regarding your model, the topology, source repository, etc., for further investigation.

 

 

Regards,

Hairul

 

Hairul_Intel
Moderator
126 Views

Hi Piyush_Khandelwal,

Thank you for your question. If you need any additional information from Intel, please submit a new question as this thread is no longer being monitored.

 

 

Regards,

Hairul


Reply