strange behavior of separable convolution in inference engine

[msong17]

I built my custom model with tensorflow and converted it to IR.

and when I tested, the output of the model changed sometimes.

I took several experiments and found out that a certain 'separable convolution' layer does not return same output even with same input

Is there anyone help me to resolve this issue?

I attached the converted model I used.

I made that separable convolution layer as output.

* I used openvino_2019.3.376, myriad stick 2

[David_C_Intel]

Hi msong17,

Thank you for reaching out.

Could you please answer the following questions:

• Which framework and topology is your model based on?
• What was the Model Optimizer command used to convert to IR format?
• Could you provide us the IR files and the frozen model?
• How are you testing your model, are you using our sample codes?
• Could you provide a sample input and expected output?

Regards,

David C.

Intel Customer Support Technician

A Contingent Worker at Intel