Extending Edge AI example app

rfronteddu · ‎2026-02-05

Hi,

I am trying to figure out what is the easiest way to extend the object detection app provided in https://github.com/STMicroelectronics/STM32N6-GettingStarted-ObjectDetection/

In particular, I would like to add one of the many N6 supported libraries such as threadx.

Lacking the ioc file, I am currently updating by hand the cubeIDE cproject file with the path to the threadx distribution I imported from the N6 examples repository but I wonder if there isn't a better way.

Streamlining or at least providing one example on how to integrate any of the provided supporting libraries would be very helpful. Can anyone point me to any documentation I can use as a reference?

Julian E. · ‎2026-02-20

@rfronteddu,

Thank you very much for this detailed answer. I saw the same shortcomings regarding our documentation. Seeing that you pointed the same issues is great as it means we are currently working on solving the right issues.

@rfronteddu wrote:
I think from a tutorial perspective ST has a few low hanging fruits such as converting a few of the more popular example projects (Object Detection, People Detection) into being generated by the new tool. For example, I think it would be very helpful to have a tutorial that goes from studio generated minimal template, to your object detection example, to threadx + led blinking example. I was able to have that working in a week so I expect someone that knows what he is doing could finish in a day or two. If that happens, please add details on the various MPU initializations etc. I think that would shed a lot of very generalizable knowledge.

For this, currently STM32CubeAI Studio generate only one "validation project" but we are thinking on providing more. I will share this idea.

For the details on initializations etc, we are writing a tutorial to explain how this "validation project" is done.
I feel that before in X Cube AI, you were given a project with no explanation, so I want to avoid it going further.

I was more focused on baremetal tutorials for now, but I think you are right to mention the use of OS. I take note of it.

@rfronteddu wrote:
Another minimal example that would be very useful to me would be a project setup to run two networks. I have seen the documentation but I am sure I will encounter hurdles when I try in a couple of months so a live example would be quite useful.

This is also something I asked.

Currently STM32Cube AI Studio do not support it, so in the very list we plan to write a tutorial.

Probably to explain how to add a NN to the "validate project".

Later, it will probably be supported directly in the tool.

The idea with the previous point is to start from nothing -> understand how to produce something similar to the "validate project" -> then improve upon it (like add a NN or explain how to produce something similar to the getting started)

Anyway, that at least the plan for now.

@rfronteddu wrote:
Regarding the WIKI, I find it very useful, I think there are some inconsistencies with the current state of the ai toolkit (for example it is recommended to use a method with threadx that is already used by default without it) but nothing too mischievous. It would be very useful if you added a search function because as of now I have to go back to google and basically google myself inside the wiki.

I would also consider adding a tool for registered users to either contribute or report mistakes. Another cool thing that I saw in the past was to have the wiki separated by revision so that if there is a breaking update you have separate valid documentation for each.

For the versioning, you can find the same doc on the ST Dev Cloud site:

https://stedgeai-dc.st.com/documentation

(I don't understand the difference between what you linked and what I linked honestly).

For midterm, we are working on regrouping all the AI docs and tutorial in one place (with a search).

But we are still doing some tests, managing multiples versions of tools with links between them is the trickiest part.

We are starting to work on it, so it may take a bit of time to be public, but we should only be better from now on.

Have a good day,

Julian

In order to give better visibility on the answered topics, please click on 'Accept as Solution' on the reply which solved your issue or answered your question.

View solution in original post

Julian E. · ‎2026-02-20

Hi @rfronteddu,

There are indeed no ioc in the getting started.

My understanding is that the N6 was not well supported by X Cube AI, so the application team did not use cubeMX.

We deployed a new tool to replace X Cube AI. This tool is a desktop app called STM32Cube AI Studio:

https://community.st.com/t5/developer-news/introducing-stm32cubeai-studio/ba-p/876445

Here you can generate a simple validation application with cubeIDE and have a .ioc .

We acknowledge the lack of documentation. We now have one person dedicated to that.

We are planning for now to add tutorial to explain how to manually add a library to a project or said differently, explain how the generated project in STM32CubeAI Studio are made. Hopefully this will help you.

Also, you have some documentation here that could help you in the meantime:

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

Have a good day,

Julian

In order to give better visibility on the answered topics, please click on 'Accept as Solution' on the reply which solved your issue or answered your question.

rfronteddu · ‎2026-02-20

Thank you @Julian E.

That is indeed great news! I was manually importing the other middleware I needed and it is a very tedious process. I will be trying the tool when I port my prototype to the Q board from the -DK.

I think from a tutorial perspective ST has a few low hanging fruits such as converting a few of the more popular example projects (Object Detection, People Detection) into being generated by the new tool. For example, I think it would be very helpful to have a tutorial that goes from studio generated minimal template, to your object detection example, to threadx + led blinking example. I was able to have that working in a week so I expect someone that knows what he is doing could finish in a day or two. If that happens, please add details on the various MPU initializations etc. I think that would shed a lot of very generalizable knowledge.

Another minimal example that would be very useful to me would be a project setup to run two networks. I have seen the documentation but I am sure I will encounter hurdles when I try in a couple of months so a live example would be quite useful.

Regarding the WIKI, I find it very useful, I think there are some inconsistencies with the current state of the ai toolkit (for example it is recommended to use a method with threadx that is already used by default without it) but nothing too mischievous. It would be very useful if you added a search function because as of now I have to go back to google and basically google myself inside the wiki.

I would also consider adding a tool for registered users to either contribute or report mistakes. Another cool thing that I saw in the past was to have the wiki separated by revision so that if there is a breaking update you have separate valid documentation for each.

Sorry for the long post! I am very excited on working on these tools so I appreciate all the work your team and ST has been putting into this.

Julian E. · ‎2026-02-20

@rfronteddu,

Thank you very much for this detailed answer. I saw the same shortcomings regarding our documentation. Seeing that you pointed the same issues is great as it means we are currently working on solving the right issues.

@rfronteddu wrote:
I think from a tutorial perspective ST has a few low hanging fruits such as converting a few of the more popular example projects (Object Detection, People Detection) into being generated by the new tool. For example, I think it would be very helpful to have a tutorial that goes from studio generated minimal template, to your object detection example, to threadx + led blinking example. I was able to have that working in a week so I expect someone that knows what he is doing could finish in a day or two. If that happens, please add details on the various MPU initializations etc. I think that would shed a lot of very generalizable knowledge.

For this, currently STM32CubeAI Studio generate only one "validation project" but we are thinking on providing more. I will share this idea.

For the details on initializations etc, we are writing a tutorial to explain how this "validation project" is done.
I feel that before in X Cube AI, you were given a project with no explanation, so I want to avoid it going further.

I was more focused on baremetal tutorials for now, but I think you are right to mention the use of OS. I take note of it.

@rfronteddu wrote:
Another minimal example that would be very useful to me would be a project setup to run two networks. I have seen the documentation but I am sure I will encounter hurdles when I try in a couple of months so a live example would be quite useful.

This is also something I asked.

Currently STM32Cube AI Studio do not support it, so in the very list we plan to write a tutorial.

Probably to explain how to add a NN to the "validate project".

Later, it will probably be supported directly in the tool.

The idea with the previous point is to start from nothing -> understand how to produce something similar to the "validate project" -> then improve upon it (like add a NN or explain how to produce something similar to the getting started)

Anyway, that at least the plan for now.

@rfronteddu wrote:
Regarding the WIKI, I find it very useful, I think there are some inconsistencies with the current state of the ai toolkit (for example it is recommended to use a method with threadx that is already used by default without it) but nothing too mischievous. It would be very useful if you added a search function because as of now I have to go back to google and basically google myself inside the wiki.

I would also consider adding a tool for registered users to either contribute or report mistakes. Another cool thing that I saw in the past was to have the wiki separated by revision so that if there is a breaking update you have separate valid documentation for each.

For the versioning, you can find the same doc on the ST Dev Cloud site:

https://stedgeai-dc.st.com/documentation

(I don't understand the difference between what you linked and what I linked honestly).

For midterm, we are working on regrouping all the AI docs and tutorial in one place (with a search).

But we are still doing some tests, managing multiples versions of tools with links between them is the trickiest part.

We are starting to work on it, so it may take a bit of time to be public, but we should only be better from now on.

Have a good day,

Julian

In order to give better visibility on the answered topics, please click on 'Accept as Solution' on the reply which solved your issue or answered your question.