Deploying a deep learning model on NUCLEO-H753ZI

[Pinagadi_Meghana]

Can anyone please help with the step by step process for deploying a deep learning model on NUCLEO-H753ZI board using STM32CubeIDE and XCubeAI package...
I am not able to do it even after trying multiple times...
please help me through this .... using a video or something... as i tried many solutions here...
Or i can provide the remote access of my system to you... if it is possible... 
It is really urgent and needed... 

[Julian E.]

Hello @Pinagadi_Meghana ,

 

Can you share your model?

Based on your previous thread, I would say that it is your tools that have an issue. 

Have you tried reinstalling them or using another pc?

 

Have a good day,

Julian

[Pinagadi_Meghana]

Actually... the same thing tried in another system also giving the same error...
I think it is not about the model, because i used many models and deleted... still facing the same issue with them...
i really want to know the procedure... 
can you please explain taking any example through a video or something 

[Julian E.]

Here is what to do exactly (I have a STM32H743)

 

Connect your board using a data cable.

You need to plug it in the ST link port (should be on top for you)

Open cube MX and select your board (I click ok for everything asked)

Make sure to have installed H7 software package and X Cube AI software package.

 

Activate X Cube AI and select Validate template

Then select onnx, tf or tflite based on your model

Add network and Click validate on target

Port COM should be found in Auto but you can check selecting it manually.

(I think that if you only see one COMx port, your cable is not data)

Check the box to enable the automatic deployment

Selec the ST Link and your tool Chain. I use CubeIDE

 

Click validate.

Here is what I get after doing exacly that:

 

 

I attached a simple Cnn model .onnx if you want to try it.

 

If this does not work, there is an issue with your CUBEMX or CUBEIDE. I would suggest uninstalling everything and download everything again. Even better would be to use another PC.

Make sure you have a data cable. It should be quite thick.

 

Have a good day,

Julian

[Julian E.]

Hello @Pinagadi_Meghana ,

It blocks at 70%, I wait a little bit after that and close it. I don't know if waiting helps, that just what I did and it worked.

Julian

[Pinagadi_Meghana]

I did that too.... but ended with the same error

[Julian E.]

It is very strange...

Can you try to use the dev Cloud instead: Session - ST Edge AI Developer Cloud

 

• Load you model to the dev Cloud
• Select the STM32 MCU as platform
• Quantize it if not already done and optimize it following the step
• Click to "Go to benchmark"

First try to benchmark it on the STM32h7 Disco or any other powerful board) and tell me if you get an estimation of the inference time.

 

Then in the last step,

• download the .elf of the project
• Select you board
• flash it to your board with STM32CubeProgrammer and open a serial terminal with a baudrate of 115200 to see if you get metrics via serial of your model running on the board.

 

If this works, you can download the MX or IDE project to have the sources instead of the .elf firmarwe.

 

Julian

[Pinagadi_Meghana]

If u dont mind.... please help me with this doubt...

After downloading .elf file... how to open that in IDE ....

Can u please send screenshots from start to end(like u have done before).... please....

[Julian E.]

Hello @Pinagadi_Meghana ,

 

You need to open it with STM32CubeProgrammer:

1. Connect the board to the pc and click connect in STM32CubeProgrammer
2. Go to the left and click the second icon
3. Browse for your .elf
4. Click Start programming

 

Then if you open Tera Term for example, select the right COM Port and set the baudrate to 115200, you should have the results of inference with a random input:

 

Have a good day,

Julian