Error during quantization

[dopeuser]

I am trying to convert the inception model to Int8 using the post optimization tool. I have the intermediate model (.xml and .bin) files. I tested this model using the inference engine and it works. But when I try to do quantization, I get the following error.

"KeyError: 'average_pooling2d_7/AvgPool'".

I am not sure how to proceed further.

I am using openvino version 2020.3.194

[Iffa_Intel · Accepted solution]

Greetings,

Please note that in the latest OpenVino version , INT8 quantization calibration is not available in the following cases:

• your project uses a generated dataset
• your project uses a model with Intermediate Representation (IR) versions lower than 10
• your model is already calibrated
• you run the project on a Intel® Processor Graphics, Intel® Movidius™ Neural Compute Stick 2, or Intel® Vision Accelerator Design with Intel® Movidius™ VPUs plugin

Please help to carefully consider above exceptions and follow this latest documentation:

https://docs.openvinotoolkit.org/latest/_docs_Workbench_DG_Int_8_Quantization.html

Sincerely,

Iffa

[dopeuser]

1. I am not sure what you mean by generated dataset? I am using a basic cats vs dogs dataset.

2. My IR version is 10.

3. It is not already calibrated as the .xml file shows fp32 as element type.

4. I have an Intel Processor Graphics chip but I don't understand how it should affect conversion of the model.

[Iffa_Intel]

Greetings,

Generated dataset means any other datasets than these 5:

• ImageNet
• Pascal Visual Object Classes (Pascal VOC)
• Common Objects in Context (COCO)
• Common Semantic Segmentation
• Unannotated dataset

Above is the list of dataset types available to use in the DL Workbench.

Your dataset does not need to contain images from official databases providing these types, like ImageNet or Pascal VOC, but it needs to adhere to the supported dataset formats

You can refer here for further infos:

https://docs.openvinotoolkit.org/latest/_docs_Workbench_DG_Dataset_Types.html

Sincerely,

Iffa