Train the openvino model of patchcore

[ShawnGao123]

When I trained patchcore model, I want to get the openvino model, so in the config.yaml of patchcore, I set

optimization:
export_mode: openvino

But at the end of training, there is always the RuntimeError:
Exporting the operator cdist to ONNX opset version 11 is not supported.
Please feel free to request support or submit a pull request on PyTorch GitHub.

How can I solve this problem?

Thanks very much.

[ashwinvaidya17]

@ShawnGao123 This issue (#440) might address your problem. The exact fix depends on your pytorch version

[ShawnGao123]

You mean the CPU model of my device? It is Intel(R) Core(TM) i7-10710U CPU @1.10GH

I use the openvino model of padim to predict some images on the my machine with above CPU (without GPU), it is successful.

But I also want to inference the openvino model of patchcore on the same machine, because patchcore have better performance than padim.

Thanks.

[ilyachur]

@ShawnGao123 Does machine with GPU have another CPU model? Or are CPUs the same?
@dmitry-gorokhov Do you have thoughts?

[ShawnGao123]

Yes, the machine with GPU has the CPU Intel(R) W-2140B [email protected] GHz

[dmitry-gorokhov]

@ShawnGao123 In most cases such error means there is no enough memory (DDR) to infer this model. What DDR capacity you have on Intel(R) Core(TM) i7-10710U? If Intel(R) W-2140B has bigger amount of memory could you please try to run OpenVINO on this CPU?

[ShawnGao123]

You are right. The machine with GPU also have a better CPU DDR capacity than the machin only with CPU.