x-cube-n6-ai-people-detection-tracking model generated binary file doesn't match

[mountzhang]

I used "generate-n6-model.sh" script to generate "network_data.hex" file as in the repo. However, it didn't match the original file. As a result, when I programed it to dev board "STM32N6570-DK", the app only shows images, and didn't detect persons. On the other hand, if I programed the original "network_data.hex" to the board, the app did detect persons. It seemed the "st_yolo_x_nano_480_1.0_0.25_3_st_int8.tflite" file didn't match the generated hex file.

[Billy OWEN]

Hi @mountzhang 

This post has been escalated to the ST Online Support Team for additional assistance.  We'll contact you directly.

Regards,

Billy

[DavidNgan]

I encounter the same issue. Can I also have the Online Support Team for additional assistance? Or is there a solution to this issue?

[Julian E.]

Hello @DavidNgan,

I suggest using a git bash. If you do the command described in the getting started with a CMD, the copy command does not provide the same result as the cp in git bash...

I encountered this issue in the past.

Else I let @Billy OWEN tell you if they found the cause of the issue.

Have a good day,

Julian

[mountzhang]

With updated of v2.1.0, I now saw hex file generated from "st_yolo_x_nano_480_1.0_0.25_3_st_int8.tflite" matched "network_data.hex".

[Julian E.]

Hello @mountzhang,

The version of the getting started depends on the version of the st edge ai core.

You need to be careful and use the right version of the Core as there are some differences, which can indeed explain the difference you observed.

Have a good day,
Julian

[mountzhang · Accepted solution]

Good to know, I will keep that in mind. Meanwhile, I was able to retrain the model, regenerate the hex file and detect object of interest (that would be separate topic). Thanks for updating the repo!

[DavidNgan]

The issue still occurs after I have changed to use Git Bash instead of PowerShell.

I am following the steps from the STM32N6-GettingStarted-ObjectDetection, link: https://github.com/STMicroelectronics/STM32N6-GettingStarted-ObjectDetection/tree/main?tab=readme-ov-file#tools-version.

The tool versions I am using are as follows:

• STM32CubeIDE (v1.19.0)
• STM32CubeProgrammer (v2.20.0)
• STEdgeAI (v2.2.0)

[DavidNgan]

I changed to test the following:

• x-cube-n6-ai-people-detection-tracking
• x-cube-n6-ai-multi-pose-estimation

Both work perfectly! Both the generated network_data.hex file and network_data.hex from Binary folder work.

[Julian E.]

Hello @DavidNgan,

So, you still have an issue with the object detection one right?

Could you please make sure that you have the latest version of this getting started (clone it again) since you are using the ST Edge AI Core 2.2.

You confirm that you try to deploy this model right: st_yolo_x_nano_480_1.0_0.25_3_st_int8.tflite

Have a good day,

Julian