Question/Issue:
I downloaded the ST-Neural ART library for my model trained on Edge Impulse. I’m using a YOLOv5 model (EI says it doesn’t support downloading the CUBE.mx CMSIS format for that model.) Do you have any documentations or guides showing how to pull this library into a STM32CubeIDE project? The only guide I could find is the CUBE.mx CMSIS guide and that is not applicable to the ST-Neural ART format. I want to be able to call the run_classifier from within my project, currently, with what I’ve tried, I can’t get the code to build (lots of path/dependency issues).
Project ID:
[Provide the project ID]
Context/Use case:
Object detection
Steps Taken:
Created model and downloaded ST-Neural ART library
Tried to import into STM32CubeIDE, but can’t get it to work
Expected Outcome:
Being able to call the functions for the model within STM32CubeIDE.
Actual Outcome:
Build and path issues. Reproducibility:
You have the right path here by exporting as ST-Neural ART library but let me check the recommended IDE and next steps with embedded team for you to get the inference working. I’m not sure why it wouldn’t allow export of CUBE.mx CMSIS for that board.
Quick follow up with steps from @ei_francesco on our embedded team:
The current recommended path is to make your changes in the network.c of our public firmware and then follow the remaining steps in our firmware quickstart guide [here] and you should be able to import as a project (GitHub - edgeimpulse/firmware-st-stm32n6)
Open your project deployment EI ST‑Neural ART zip
You’ll find files like:
network.c(this is what you need to replace @Williedlbat)
network_data.hex (the compiled network blob for the NPU)
Replace network.c in your firmware project
In your CubeIDE project (or the reference FW), locate the existing network.c and replace it with the one from the EI export zip.
Let us know how the project works for you or if encounter any further path / dependency issues please share the IDE version, Development OS and project ID so we can reproduce with the team.
Hope this helps, and thanks again @ei_francesco for the advice.