Edge AI

Dear Sir/Madam, I am getting an error "Analyze fail on the AI model" when I analyze a network using "STM32Cube.AI runtime" (see the snapshot below). I just train a simple AI model (https://colab.research.google.com/drive/1HBsuTW2gGAoPgWnkpwXiXI7JHXR05lrG?usp=sharing#scrollTo=eFymXEWBwarM) and want to execute a model (sine_model.tflite) on the board. However, when I want to analyze the network, I get this error "Analyze fail on the AI model".I am using the latest version (1.12.1) of STM32CubeIDE and X-Cube-AI(8.0.1). I work on Apple M1 Pro (Macbook Pro). Could you please provide a solution so that I can analyze the network on STM32CubeIDE?

..

Hi,I am using the X-Cube-AI expansion pack in STM32Cube IDE and wanted to add an ONNX model. After selecting the model and pressing Analyze, I get the error: "TOOL ERROR: Applying softmax on single element".The model I am trying to use is a neural network created in Scikit-learn and is converted into the ONNX format using skl2onnx. I am using skl2onnx version 1.14.1 and X-Cube-AI version 8.0.1. The project was created for the SensorTile.box.Although I am trying to create a neural network for my own data, I am receiving the same error message when using the example for creating a neural network and converting it to the ONNX format. The following is the example code I used in Python: # Train a model. from sklearn.datasets import load_iris from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestClassifier iris = load_iris() X, y = iris.data, iris.target X_train, X_test, y_train, y_test = train_test_split(X, y) from sklear

Hello!I'm getting the error: INTERNAL ERROR: list index out of range when uploading an onnx model to STM32Cube.AI and run analysis. I have validated my model with other tools like netron.app and it is correct. I would appreciate if anyone could help to figure out what could be the problem? I attach my model below in case someone can help me. Thanks in advance. Regards!

hi,When I analyze the model, I always get an error.How should I solve it? Thanks!

The Data Logger does not light up yellow as shown in the figure below after starting Nano Edge AI Studio.Also, when I try to create a model for anomaly detection, I cannot select IMPORT from [Type of signal sourse] as shown in the figure below, and I cannot proceed.Can you please tell me if you can think of any possible causes? 

..

Good morning,I'm having some issues with the MLC configuration of a LSM6DSV16BX sensor through Unico GUI.I've followed the tutorials provided by ST online and I've looked through the related GitHub repository, but, when I save a datalog through Unico GUI, the .txt file seems to have an additional first column of data called "Comm_Interface", which takes the value "UI" in each row.When I try to upload the data to the MLC section of unico, all the values are seen as 0: by manually eliminating the first column (which is not present in any of the examples that were provided), the values are interpreted correctly.Is there a reason why such column is added to the .txt file? I couldn't find any examples with this same issue and I'd like to know if the file can be saved correctly, thus removing the need to modify it before using it (especially since I'm using the same platform to log and analyze the data).Thanks in advance for the help!

I have uploaded the tflite model. I am using an input array which is basically an image converted to a byte array and I want to classify this input array into one of the two classes.Please help. I am using ver 7.1.2 of X-CUBE-AI

..

Here are the software versions I'm using:STM32CubeMx - 6.3.0MCU package for STM32H7 - 1.9.0X-CUBE-AI - 7.0.0MCU - STM32H743ZITxIAR EW for ARM - 8.40.1I run into this problem since I have added the X-CUBE-AI software pack. every time I open the X-CUBE-AI configuration window, I get the following error:Clock and Peripheral parameters not set for best performance, automatically fix?further inspection shows that the following configurations are modified every time I open the X-CUBE-AI configuration window:System_Core->RCC->HSE: changed from "Crystal/Ceramic Resonator" to "BYPASS Clock Source".Clock_Configuration->HSE_Input_Frequency: changed from 16 to 8.System_Core->CORTEX_M7->CPU_I_Cache and CPU_D_Cache are FORCED to "Enabled" and cannot be disabled.IAR->Project->Options->C/C++_Compiler->Language_1->Language is changed from "Auto" to "C".This is extremely inconvenient and make experimenting with neural network very frustrating and impractical. Please fix / p

After installing the latest version of CubeIDE, there was a problem with the Nucleo F746ZG projects. It turned out that the default configuration of peripherals (especially ethernet) causes errors. The printf function worked badly (e.g. it could not cope with the delay function), and after using the X-CUBE-AI plugin, it stopped working at all. Only installing older versions fixed these problems (CubeIDE 1.0.2; CubeMX 5.3.0; MCU F7 1.15.0; X-CUBE-AI 5.1.2). Are you also having problems with the new versions, or am I doing something wrong? I don't know how the intermediate versions behave, previously I used Eclipse in CubeMX and SW4STM32 plugins and everything worked.

Hello!I'm getting the error: INTERNAL ERROR: list index out of range when uploading an onnx model to STM32Cube.AI and run analysis. I would appreciate if anyone could help to figure out what could be the problem? I attach my model below in case someone can help me. Thanks in advance. Regards!

I know that the user has to set a window lenght (for example 104 samples). When you deploy the created algorythm, this window takes jumps the size of the window lenght or does it move sample by sample overlapping some values?

..

Hi,I'm trying to implement a detection model on a STM32-H747i-disco board, the only example provided is the one called "people counting", but the libraries are provided in binary format and it seems that it is not possible to modify them. How could we implement a detection model to detect specific objects (such as cars) ? Or is it possible to fine tune an existing detection model ? Thank you for your help,R.SOLER ALTEN

Hi, I recently purchase a STM32H747 that have a touch display. The problem is that every example in the FP-AI-VISION that have to display the B-CAMS-OMV image on the screen doesn't work. I tested the example that use the STM as a webcam and it work's. I tested some graphical examples for the STM32H747 and some do not work. In the examples that display the camera on the screen I have the same issue as in the FP-AI-VISION examples. There are grafical examples that does not involve the camara and it happens the same. The LCD display this:

Hi,I'm trying to do the "validation on target" of a neural network using X-CUBE-AI on STM32CubeIDE on Windows 11. I'm using STM32F401RET6. The "validation on desktop" works fine but the "validation on target" gives me the error in the attached picture.I tried to update the STM32 Virtual COM Port driver but with no results. This error is given only when communicating by USART, if I use ST-Link (i.e with STM32CubeProgrammer) I can read all the registers so I think it works fine.Thanks.

..

..

The error is in the network.c code file generated by the X-CUBE-AI library. I have chosen "Not Selected" option in drop down option for X-CUBE-AI. The code below is giving the error as mentioned:

Dear ST Community Support,I hope this email finds you well. My team and I have been working with theSTEVAL-PROTEUS1 and have successfully utilized the STMicroelectronics NanoEdge AI Studio to create an Anomaly Detection (AD) model for our project. As a result, we have obtained the .a and .h files, as well as the sample app main code.We are now in the process of integrating the AI library into our application, and we require some assistance. In the provided sample app main.c code, there is a function called fill_buffer(input_user_buffer) that seems to be responsible for filling the buffer with sensor data. Unfortunately, we could not locate the source code for this function.We kindly request your assistance in providing the source code for the fill_buffer(input_user_buffer) function, as this will help us understand how to properly manage and process the sensor data in our application. This information will be invaluable in ensuring that our deployment proceeds smoothly.Thank you in adva