Edge AI

Hi, Thanks for your attention. I am analyzing the .onnx format model but an error occurs. I try to reinstall the stm32cubeMX but it doesn't help. The part of architecture is: Here is the log information. Can you help me with this problem? @Romain LE DONGE​  @jean-michel.d​ @fauvarque.daniel​  Thanks very much!2022-07-08 15:39:51,314 [INFO] BoardInfo:879 - No configuration file found for board P-NUCLEO-WB55 2022-07-08 15:39:51,316 [INFO] DbBoards:156 - Kit is not supported: P-NUCLEO-WB55 2022-07-08 15:39:51,319 [INFO] BoardInfo:879 - No configuration file found for board STEVAL-BFA001V1B 2022-07-08 15:39:51,320 [INFO] DbBoards:156 - Kit is not supported: STEVAL-BFA001V1B 2022-07-08 15:39:51,321 [INFO] BoardInfo:879 - No configuration file found for board STEVAL-BFA001V2B 2022-07-08 15:39:51,321 [INFO] DbBoards:156 - Kit is not supported: STEVAL-BFA001V2B 2022-07-08 15:39:51,404 [INFO] DbBoards:160 - Found 176 boards, 173 are supported 2022-07-08 15:39:51,408 [INFO] A

Hi ! I perform predictive maintenance on electricals fans with steval_stwinbx1. I created the IA nCC library using the NanoEdge IA software. Then I added the library in the Steval-stwbx1 IDE project as indicated in the link  Then I flashed the binary file, created after building the IDE project, via the STM32CubeProgrammer software using USB mode. So I consider that at this stage, the IA library has been integrated into the steval-stwbx1 microcontroller and is ready to detect anomalies. I've downloaded Tera Term but when I connect the steval-stwbx1 to the computer. I've downloaded Tera Term but when I connect the steval-stwbx1 to the computer via the usb-c cable, Tera Term doesn't detect it as a connected device. I've tried configuring the card's usb to detect but I don't know how to do it. So my question is: do I need to configure the card's usb? If so, how do I do it? another thing: also on the NanoEdge software, the usb port is not detected for direct data acquisition. Please s

Hi,I'm trying to test X-CUBE-AI on my Stm32F429ZI nucleo board. I'm using Stm32CubeMX, Stm32CubeIDE.Setup: stmnist_v1.h5, I've PNG files for this dataset.At first I'd like to test the AI routines with in-memory buffer so I'd like to know what is the input buffer format. So when I have a BMP or an PNG file how to extract to the memory.  

Hi team. May you kindly assist how to eliminate false positive for a face recognition project  is it my code which is the problem or something else. For white people I am getting true positives but for blacks, only false positives.

Hi,I am currently using LSM6DSOX and its features like MLC and built in pedometer in a project, MLC works just fine but when I am walking whilts LSM6DSOX is in a 40 to 50 degrees tilted position on my wrist, pedometer misses around half of the steps taken. I tested this using a stm32 devkit and i didn't have this issue. I didn't see anything on the datasheet that says the position of the IMU sensors affect the pedometer performance either. Any ideas what might be the problem?

I've updated my code to CubeAI 8.1. The middleware is still deleted... but I can add it manually. However when I build the code I gett the error:Description Resource Path Location Typecannot find -l:NetworkRuntime730_CM7_GCC.a AI_FILES_GENERATOR C/C++ ProblemHowever, the file is there in Middlewares and the library paths have been generated. I have even added a library path to where this file is located and still it does not work.I have created a new project with STM32CubeIDE with the latest CubeAI and the problem is still there.Tool settings>LibrariesError:File in the project:  Thank you for any help 

Hello,Can we generate analog output from MLC instead of a statement? If we can, could you explain how to do it? I plan to use MLC output as a part of my control algorithm. What is the most efficient way to achieve this?

Hello good afternoon, I am starting to make an application with the sensortile.box very similar to "baby crying". I would like to ask if you have any information that can support me in this process, for example to activate the use of the AI folders in the IDE. If you have any more detailed information on the use of AI in tile.box, it would be very helpful.Thank you very much.

我真的不知道为什么..将出现此问题。将 YOLOv5 转换为 ONNX 后，我无法在其上部署它，因为我的显卡是 1660ti 并将半精度转换为单精度。如果我使用半精度，它可能不支持 16 位半精度（所以要减小我的权重文件的大小并不容易），但如果我切换到单精度，则意味着列表溢出，并且没有指示溢出发生的位置。我真的很相信，事实上，我还没有看到一些 Yolov5 部署在 cubeAI 上......一些大人物知道为什么在这里

Hello every Body, i am investigating the use of ISM330IS with nanoEdgeAI to detect Anomaly situations. Lets take a motor for air conditioning as an example, If the motro  has protection against high current, and a tachometer, What are the benefits of ISPU  against those typical protection.the seconde question is ,is there any industrial project using the ISPU to detect anomality ? thank you very much

Hi, I am trying to use a QKeras model with mixed precision on STM32H747I-DISCO board. I run ‘Analyze’ my QKeras model using stmai 8.1.0, but encountered the following error. I tried an old version of QKeras, but got a different error. I would appreciate any help on how to solve this issue?Thanks in advance! ------------- error ------------- >>> stm32ai validate --model model_test.h5 --allocate-inputs --allocate-outputs --relocatable --compression none --optimization balanced --name network --workspace workspace --output outputNeural Network Tools for STM32 family v1.7.0 (stm.ai v8.1.0-19520)E010(InvalidModelError): Couldn't load Keras model model_test.h5,error: Error when deserializing class 'QBatchNormalization' using config={'name': 'bn_0', 'trainable': True, 'dtype': 'float32', 'axis': [3], 'momentum': 0.99, 'epsilon': 0.001, 'center': True, 'scale': True, 'beta_initializer': {'class_name': 'Zeros', 'config': {}, 'shared_object_id': 6}, 'gamma_initialize

What I have checked through the user manual UM2526 and FP-AI-VISION1 examples is that I am using X-cube-ai version 6.0.0 for the 32bit floating point model (the model data type I am currently using), so I want to read data from external SDRAM in order to reduce inference time. I know that it is optimized so that the inference time is faster than reading from the internal flash memory. Currently, the model I trained is Keras 2.3.0, which is not compatible with X-cube-ai 5.0.0.Purpose : To reduce the inference time, I want to save the initial weight&bias table to the external Q-SPI flash memory and move it to SDRAM as in the example. Question 1 : The inference time is faster when it is placed in the external memory when the data type is float, but in the table12, internal flash memory shows faster results , I don’t understand why. (reference : UM2611 manual)Q2 : In general, what clock delay is the read time latency of the internal flash memory?Q3 : If I use external SDRAM for re

Hello STM Community,I am currently trying to upload a large model to my STM32H745I-DISCO. Given the size of the model, I chose the [Use external flash] option, which, as per the user manual, led to the generation of a separate network_data.bin file. As per my understanding, this network_data.bin file should contain the weight files that were split, although I invite corrections if my assumption is incorrect.The user manual further emphasizes that the network_data.bin file must be manually programmed on the board's external Flash memory. Tools such as the STM32CubeProgrammer (STM32CubeProg) are recommended for this task. An interesting note I came across was that using automatic validation on the target leads to the network_data.bin file being automatically programmed on the external Flash memory on the board.Given this information, I'm left with some uncertainty: if I don't manually code to explicitly utilize this network_data.bin file, will the model still be able to run inferenc

 Hi STM Community,I'm currently leveraging an STM32H745I-DISCO to run our custom DNN model, which is designed for a 4-class classification task and accepts multivariate time-series data of the shape (3, 2048). While I've successfully analyzed and validated the model with random input data on the target platform, I'm finding it challenging to upload and run inference using our own custom data (over 100 instances).Considering my limited experience in embedded engineering, I'm in need of detailed guidance on how to accomplish the following:Data Uploading: What are the steps and the preferred format to upload our custom data onto the STM32 platform?Data Implementation: How can I integrate this data into the main script for the model to run inference?Data Reshaping: The first reference below seems to illustrate the use of 1-dimensional data, while our model uses 2-dimensional data. Could you provide advice on reshaping the data appropriately for our model?Here are a few resources I hav

Hello! Im using STM32Cube IDE version 1.12.1.Im having isssue with including the header file "ai_platform.h " and "ai_datatypes_defines.h".I have installed X-cube-AI extensions  and all other required extension of Artificial Intelligence.I am getting the error message of "no such file or directory" .I request you to look into the issue.Thank you !

Good Morning , I am working on the following board:  https://www.st.com/en/evaluation-tools/stm32wb5mm-dk.htmlI need to activate Unicleo-GUI and MLC core on the board.  Do you know how to get it ?Thank you very much. Best regardsLuigi Bibbo'PS: my email is luigi.bibbo@unifi.it

How to deal with fatal error: ai_platform.h: No such file or directory? I can't find this file anywhere in the project.I tried using different versions of CubeIDE and X-CUBE-AI, added include paths in project properties. I checked on F7 and H7. I have Core and ApplicationTemplate selected in Software Packs Component Selector.The problem started with the inability to use USART along with the X-CUBE-AI package enabled. Then no error appeared, only communication over USART did not work. I am currently using CubeIDE 1.12 and X-CUBE-AI 8.0, and have updated what I can. The last time AI worked for me I was still using Eclipse (SW4STM32), but it no longer supports the latest chips I use.Please point out what could be the problem.

If i want to send the result of (Food-Recognition)of FP-AI-VISION1_V3.1.0,to achieve to object of  to communicate some event that the network recognizes (for example the detection of a certain object) and send it out of the panel.The complete project is shown: development board + external board (raspberry or arduino or whatever).Hay some software o method o project to achieve this?

Hi I have a really simple Neural Network (Keras) with 3 float32 inputs and 3 float32 outputs. Size: 2.2KB RAM 16kBFlash and 1600MACC, really small network, in my opinion it should not be a problem at all to run that.For testing i wanted to try the simplest way possible to run that Network on a STML4 MCU.I used CubeMX, integraded the Model and created and built an example Project by using a "ApplicationTemplate".All went prefect. so know i have an example Project where everything is generated. especially a file called "app_x_cube-ai.c".There is a init function called: "MX_X_CUBE_AI_Init()" and a RunAI function called " MX_X_CUBE_AI_Process", looking like this void MX_X_CUBE_AI_Process(void){    /* USER CODE BEGIN 6 */  int res = -1;   printf("TEMPLATE - run - main loop\r\n");   if (network) {   /* 1 - acquire and pre-process input data */  res = acquire_and_process_data(data_ins);  /* 2 - process the data -

Hello everyone!At the moment I am developing an AI system for the STM32H7A3 using STM32 Model Zoo und Cube AI. For testing purposes I trained the SSD Mobilenet V1 model for 192x192 picture resolution and implemented it in a firmware app for testing interference. For training I followed the tutorial in github.The output of an interference is of the shape 3830x29. Where can I find out what that means and how can I calculate detected class, probabilty and boxes of detection?If you need further informations just ask.Anybody can help me? The info in the github just says "Coming soon".

I'm trying to do a program like a baby crying and i'm looking for the code of babycrying program. I need to do a program that detects unwanted noises in cars, so i think that maybe is similar than the babycrying app.I thought it was possible using the expert mode, but now I think it is not possible with it. If someone could help me pls, maybe with some information or the babycrying code I would appreciate it.

Hello everybody,I started to use the AI Cube MX extension and developed an application on the NUCLEO-F746ZG Board. It all worked very well.Now I am trying to integrate this small application into a bigger, existing project.There seems to be a problem with the Init function of the AI, it causes a reset after a few seconds.There are a  few possible problems I identified:- the project was configured with a CubeMX Version the AI extension is not compatible with- the clock configurations of the two project parts seem to clash (HSE_ON vs. HSE_BYPASS)Does anyone have experience with that?Does anyone know any clock configuration etc. that the AI extension is compatible with? I cannot find much in the documentation...  

Good morning, I am working on this Firmware https://github.com/stm32-hotspot/STM32WB-BLE-AI-MotionSense.I created the Neural Network but I don't see the Classification on The Monitor.Can you help me to solve the problem ?Thank you.Best regardsLuigi Bibbo' PS: my e-mail is   luigi.bibbo@unifi.it 

Hi,I have managed to program the STM32F746G-DISCO with a simple neural network but when I try to do it using the external Flash by using the option in advanced settings (External NOR Flash) it doesn't work. Any idea why? Is there anything else I need to do?I'm using STM32CubeMX and Keil.Thanks!

I want to know when will stm cube ai support keras 2.12.0 because for now it only support up till 2.11.0. I can't seem to downgrade my keras version as it require to downgrade more dependencies.Furthermore I converted my lstm keras (h5) model to tflite but it gives me "transpose of batch size not supported" error on analyze any solution?