X-Cube-AI >> CNN+LSTM: NOT IMPLEMENTED: UNsupported layer types:...

[qs202310]

Dear Sir/Madam,

I am trying to deploy a trained AI model (CNN+LSTM) to STM32F767. When I tried to analyse the TFlite file with the STM32CubeMX, it stopped and gave the following message: 

Could you please help me sorting out this issue?

Thanks

QS

[Julian E. · Accepted solution]

Hello @qs202310,

Your issue here is that your model uses unsupported layers: FlexTensorListStack, WHILE, FlexTensorListReserve

The list of currently supported layers can be found here: https://stedgeai-dc.st.com/assets/embedded-docs/supported_ops_tflite.html 

We are constantly improving our tools but for the moment, you need to manually modify your network if you can or use another model.

I think that the ST Edge AI core has been updated today, you can try to use the ST Dev cloud (Session - ST Edge AI Developer Cloud) and see it works, but I don't know if the layers that you need have been implemented.

Have a good day,

Julian