Issue with deployment STM32Cube.MX

#1

Hi,

I’m trying to deploy my model on a STM32WB55CCUX using the STM32Cube.MX package, following the tutorial on the docs page here. Everything does work fine until I include the code part in the while loop:
EI_IMPULSE_ERROR res = run_classifier(&signal, &result, true);

A build / Run will then lead to the following output:

ST-LINK FW  : V2J39S7
Board       : --
Voltage     : 1.74V
SWD freq    : 4000 KHz
Connect mode: Under Reset
Reset mode  : Hardware reset
Device ID   : 0x495
Revision ID : Rev Y
Device name : STM32WB5x
Flash size  : 256 KBytes
Device type : MCU
Device CPU  : Cortex-M4
BL Version  : 0xd5
Debug in Low Power mode enabled

Memory Programming ...
Opening and parsing file: st-link_gdb_server_a13924.srec
  File          : st-link_gdb_server_a13924.srec
  Size          : 158.36 KB 
  Address       : 0x08000000 

Erasing memory corresponding to segment 0:
Erasing internal memory sectors [0 39]
Error: failed to erase memory

Error: failed to erase memory
Shutting down...
Exit.

I tried with both Quantized (int8) as Unoptimized (float32). Also played a bit witth the Linker settings Minimum Heap Size/ Stack size. Default 0x800 for both.

Any suggestions are welcome!

#2

Hello @fferket ,

Can you tell me which one of your project have the issue?
Not entirely sure it is related to your project but I want to make sure your model is not too big for your board.

Regards,

Louis

#3

Hi Louis,

This is the first project in the list of my projects (4 chars).
In this project I just made a simple model based on 3 activities (standing, walking, running) with a total of 9 samples of each 10 seconds.
Could it be possible that the BLE Stack of the STM32WB series has overlap with the EI stack?

Regards, Frank

#4

Ok thanks,

I just joined your project and will try to compile it with our C++ standalone example see if this works first.

I’ll let you know.

Regards,

Louis

#5

Ok I managed to compile and run the C++ standalone inference example on static raw features.

I will ask internally about your issue as I don’t have a straight answer now.

Regards,

Louis

#6

Thank you for your quick response and help on the compilation of the C++ standalone example.
I hope someone internally does have a solution for this, or an advice on how I can tackle this issue.

Note, I managed already to generate and import a similar .h5 model directly via STM32CubeMX (using X-CUBE-AI) for this microcontroller as an output of a tensorflow python script. But I would really prefer using your platform as this is way more intuitive.

Regards,

Frank

#7

Is it maybe possible to add an export as a .h5 model to the website?

#8

@fferket Already there, go to Dashboard > Downloads and you can find it.

#9

Hi Jan,

Thank you!
Unfortunately still having issues loading the model. The X-CUBE-AI package from ST only supports Keras models ← 2.6.0. The .h5 model is saved with Keras 2.7.0.

Since the STM32Cube.MX package also does not work for the STM32WB, might it be possible that a 2.6.0. is provided by your website?

E010(InvalidModelError): Model saved with Keras 2.7.0 but <= 2.6.0 is supported