Validation on target X Cube AI

[Talo]

Hello! I would like to implement simple Embedded NN on STM32 MCU. I have followed tutorials but still have some questions:

1. For validation on target, I have flashed hex file into micocontroller and reconnected it. When run validate on target, the following message is shown.

 Exec/report summary (validate) 
 -------------------------------------------------------------------------------------------------------------- 
 model file         :   C:\Users\Control\Documents\STM32Home\model_sine.h5                                      
 type               :   keras                                                                                   
 c_name             :   network                                                                                 
 compression        :   low                                                                                     
 options            :   allocate-inputs, allocate-outputs                                                       
 optimization       :   balanced                                                                                
 target/series      :   stm32g0                                                                                 
 workspace dir      :   C:\Users\Control\AppData\Local\Temp\mxAI_workspace1115474183200017071310444384230046    
 output dir         :   C:\Users\Control\.stm32cubemx\network_output                                            
 model_fmt          :   float                                                                                   
 model_name         :   model_sine                                                                              
 model_hash         :   0xa96cca22e011f8257d867dbaf6026b31                                                      
 params #           :   491 items (1.92 KiB)                                                                    
 -------------------------------------------------------------------------------------------------------------- 
 input 1/1          :   'input_0', f32(1x20x1), 80 Bytes, activations                                           
 output 1/1         :   'dense', f32(1x1), 4 Bytes, activations                                                 
 macc               :   9,821                                                                                   
 weights (ro)       :   2,084 B (2.04 KiB) (1 segment) / +120(+6.1%) vs float model                             
 activations (rw)   :   400 B (400 B) (1 segment) *                                                             
 ram (total)        :   400 B (400 B) = 400 + 0 + 0                                                             
 -------------------------------------------------------------------------------------------------------------- 
 (*) 'input'/'output' buffers can be used from the activations buffer 
Running the Keras model... 
Running the STM AI c-model (AI RUNNER)...(name=network, mode=TARGET) 
INTERNAL ERROR: E801(HwIOError): Invalid firmware - COM5:115200

Analyze and validation on desktop work well. My microcontroller is from G0 family, so is this a reason for this error message?

[hamitiya]

Hello,

Which hex file are you referring to ? Is your project a "validation" template ? Do you see any outputs if you open COM5 port with a tool such as PuTTY or TerraTerm ?

Best regards,

Yanis

[Talo]

@hamitiya Thank for reply. I have used .hex file which is located in folder release (I dont know if this is what you mean) and flashed it using STM32 Prog with UART port.

 Is your project a "validation" template ? 

What do you mean, sorry but I am new with STM.

Do you see any outputs if you open COM5 port with a tool such as PuTTY or TerraTerm ?

No, But accauly, I didn't modify the generated code so I don't expect any message on Putty.

[hamitiya]

Hello,

Yes the hex or elf file in Release folder.

From the documentation: Getting started with X-CUBE-AI Expansion Package for Artificial Intelligence (AI) - User manual, in page 15, you select in the wizard your project, you have 3 choices:

-  System performance: standalone AI test application for performance purpose
-  Validation: AI test application for validation purpose
- Template application: basic application template for AI application

Validation will generate a project which will allow you to validate using "validation on target". 

SystemPerformance will print you inference time directly in the output and you can read it from tty.

You don't need to modify the code to see any message on putty. X-CUBE-AI appends the project main application with either validation or SystemPerformance structure. You only need to load a model (onnx, keras, tflite)

I suggest you to follow the documentation to start with X-CUBE-AI, and append this thread if you still see issues or if you have any questions. 

Best regards,

Yanis

[Talo]

@hamitiya , Yes my project is selected from wizard for validation purposes. So Microcontroller from family G0 is suitible for AI application? It is G0B1CEU6N.

[Ritesh1]

To perform the hardware validation, first you have to flash the correct validation code. After flashing the validation code, connect the board to the Tera term. If the validation code is correct then a certain message appears on screen  (as given in the xcube ide user manual). After completing this step try the hardware validation.

Ritesh

[Talo]

@hamitiya In documentation Getting started with X-CUBE-AI Expansion Package for Artificial Intelligence (AI) - User manual

There is one option "Set the MCU memory subsystem" which is not available in my micrcontroller.

[Talo]

Putty gives some message as in image. As wriiten "Ready to receive CMD from HOST". Should I write somthing to test, but anyway the terminal (cursor) doe not respond.

[Ritesh1]

Now, try the hardware validation step.