“INTERNAL ERROR: Order of dimensions of input cannot be interpreted" anaylsing Onnx model

[Levy-hy]

Hello, I'm trying to generate C code for my Onnx model using ST Edge AI Core.

I've tried both X-CUBE-AI and ST Edge AI Developer Cloud, and they gave me the same error:

It's strange that Netron could still draw the correct graph containing the dimensions of input .Does this indicate that the input dimension has been successfully parsed?

I tried simplifying the model input, but it still reported an error. Can anyone give me some advice? Thank you.

[Julian E.]

Hello @Levy-hy,

Your onnx model seem fine as you pointed out.

The error here says that the st edge ai core (the CLI behind both X Cube AI and Dev cloud) is having an issue while converting it. It is the C conversion that has an issue, not your model.

It could come from unsupported layer, bug, other issue.

Could you share your model?

You can try to use the --use-onnx-simplifier in X Cube MX or Dev cloud.

In X Cube AI, above the Analyze button, you should see a wheel (parameters) button. In here, you can add a command (bottom of the first page menu).

Have a good day,

Julian

[Levy-hy]

Hi @Julian E. 

Thank you for explaining the error for me!

I tried adding --use-onnx-simplifier in extra command line, but then I got an error message about the transpose of the LSTM model.

Perhaps there are some layers in my model that Edge AI cannot recognize? I will try to modify the model and see if it could work.

It seems that the model type is not supported for attaching file, so I'll try to solve it myself first, thank you so much!

[Julian E.]

Hello @Levy-hy,

Yes, your error in this case seems to be due to a parameter of Transpose not being supported.

You can find the list of supported layers here:

https://stedgeai-dc.st.com/assets/embedded-docs/index.html 

In the section: Supported layers/operators

If you are using the N6 and the NPU, you have to look at the page: ST Neural-ART NPU - Supported operators and limitations.

To upload your model, you have to zip it.

Please first try to edit your model. If the issue is just that a layer is not supported, I can just report it to the dev team, nothing more.

Have a good day,

Julian