STM32N6570-DK: Custom 5-Class ST YOLOX Model Freezes After Few Frames and Unexpected LED Turns On

[Rutva Patel]

Hello ST Community,

I am working with the STM32N6570-DK and running a custom-trained ST YOLOX Nano object detection model using the STM32N6 Getting Started Object Detection application.My custom model has 5 classes.

The model is successfully generated and integrated into the STM32N6570-DK object detection project. The application starts correctly, the camera/display starts running, and I can see a few frames initially.

However, after processing a few frames, the display freezes/stops updating. The board itself does not appear to continue normal inference after this point.

I also noticed an unexpected LED indication on the board. This LED was not active when I previously ran the standard/reference object detection model. I have attached photos of the front and back of the STM32N6570-DK showing the LED state.

 

My setup is:

• Board: STM32N6570-DK
• Application: STM32N6 Getting Started Object Detection
• Model: Custom ST YOLOX Nano
• Input: 480 x 480 x 3
• Number of classes: 5
• Camera: ST camera module connected through the camera interface
• Display: STM32N6570-DK display
• IDE: STM32CubeIDE

I have attached the STM32CubeIDE screenshot showing the debugger log as well.

The main things I would like help understanding are:

1. What does the unexpected LED indication shown in the attached board photos mean?
2. Could the freeze after a few frames be caused by a HardFault, memory issue, NPU/runtime issue, cache/DMA coherency issue, or external memory access problem?
3. Is there any recommended way to capture the fault status when the application freezes?
4. Are there any additional changes required in the STM32N6 Getting Started Object Detection application when changing the reference ST YOLOX model to a custom 5-class model?
5. Which registers/logs should I inspect to determine whether the CPU or NPU is faulting when the screen freezes?
6. Could there be an issue with model input/output buffers or memory allocation that only appears after several inference frames?

The reference/example model previously ran correctly on the same hardware setup, so the problem seems to occur after integrating my custom ST YOLOX model.

I have attached:

• Photo of the front side of the STM32N6570-DK showing the LED state
• Photo of the rear side of the board
• STM32CubeIDE screenshot showing the ST-LINK/debugger logs

Any guidance on how to debug this freeze or identify the meaning of the LED indication would be appreciated.

Thanks.

 

[Julian E.]

Hi ​@Rutva Patel,

 

Here is a debug guide for the STM32N6:

How to debug STM32N6 using STM32CubeIDE | Community

 

How did you deploy your model? 

If you did it using the deploy model from model zoo, there should not be any additional steps.

If you are doing it manually, make sure to follow the instruction in the read me from the Getting started.

 

I know that for yolo models 480x480 is kind of the limit, you may try to use a smaller input.

First, don’t lose time training it, just deploy a random weights model and see if you have any improvements. 

 

 

Have a good day,

Julian