Edge AI

Hello ,I'm trying to figure out how the 1_class library works. In the Emulator tab of the NanoedgeAI Studio software, I can see that the sensor indicates that the input pattern is detected as an outlier (outside the class considered), as shown in the following image: But when embedded on microcontrollers, according to my code the orange LED should be flashing, but it's the green LED. I can't figure it out. Is my code correct? How can I retrieve the function output, which should be 0 for belongs to the class or 1 for outside of the class?Thanks for your help.A part of my code :/*________________1_class_____________________*/neai_state = neai_oneclass_init(knowledge);/* Infinite loop *//* USER CODE BEGIN WHILE */while (1){if (drdy) {/* Reset data ready variable */drdy = 0;/* Read watermark flag */iis3dwb_fifo_status_get(&dev_ctx, &fifo_status);if (fifo_status.fifo_th) {num = fifo_status.fifo_level;if (data_left > num) {data_left -= num;}else {num = data_left;data_left = 0

I have version 6.7.0 of STM32CUBEMX, version 8.1.0 of X-CUBE-AI, and my stm32 development board model stm32h743iit6. using the Analyze button on STM32CUBEMX works. But when I generate c code for my onnx model and deploy it to the development board, my model fails to initialize. When I traced how the problem happened, I saw the ai_platform_network_create function. It's a black box function and I can't get inside it (it seems to be an officially provided function). I can compile and burn normally using the Keil software, but I can't proceed further when I run into this function. When I use debugging and click on the Stop button while stuck, my Call stack+Locals window prompts an HardFault_handler error. How do I get my model to run properly?   

Hi I am trying to use the FFT function of CMSIS DSP library, along with X-Cube-Ai on Cube IDE 1.10.1As i try to use the FFT feature, while using TensorFlow lite, it gives multiple error. Is there any way to use the FFT feature with Cube Ai ?

Hi,I'm trying to analyze my Keras model (converted from Keras) in STM32 X-CUBE-AI version 8.1.0. But it shows this error: TOOL ERROR: 'gbk' code can't decode byte 0xb8 in position 577:illegal multibyte sequence.As shown in the picture.But I I'm trying to analyze in STM32 X-CUBE-AI version 7.3.0. It can be analyzed!!!Any help would be welcome!

Hello,I'm interfacing the STEVAL$MKI238AA through serial commands with STEVAL-MKI109V3. The UM2116 document in its Rev 10 - April 2022 doesn't mention any reference to the LIS2DUX12. I would like to know what is the format when I send the command *start. In table 4 (page 18) there is a long table explaining the format from all different board that can be connected but lacks of the STEVAL-MKI109V3.Could you guys provide an upgraded version of the document with this information or just the format that I'm looking for?Regards,Pascal.

Hello,I am a beginnerI downloaded the STM32-NANOEDGE-DATALOGGER firmware via the github link.This STM32-NANOEDGE-DATALOGGER is used in NaoedgeAIStudio and is streaming data via UART (VCOM port) from STLink.but I see that the example is only given for an AD (Anomaly Detection) project and I'd like to know how to modify it for an n-Class project?Also, is it possible to add the USB_CDC_ACM  class to activate the VCOM port on the STWIN.box without the STLink?If so, how do I do it?Board used: STWIN.box

Hi,I have an STM32H573I-DK Discovery kit with me. I wanted to try out Edge AI applications on it, and was checking out the STM32Cube.AI Developer Cloud. When I reached the "Generate" section and browsed through the listed boards, I could not find any of the STM32H5 series MCUs listed. Is it because they are not supported?Will I be able to use it if I use X-Cube-AI with CubeMX?

Hello,After generating my ucf file that I load to my LIS2DUX12 and verify that it works, I wanted to load the generated configuration through a microcontroller so I used the option in Unico GUI to generate the header file which later on I need to include in my project. After analyzing the content of the commands that is sending, I noticed that is sending commands to addresses not listed in the datasheet neither displayed in Unico GUI. This concerns me as these could be reserved registers and then damage the accelerometer:Reviewing the datasheet, the highest address described is 0xB7:Could you point me what is then wrong? Is it the documentation and Unico GUI or the generated file?

Hi, @Denise SANFILIPPO , I checked this post and was wondering how to check whether my IMU has sensor fusion functionality? It seems that my adapter is also MKI239AA, looking exactly the same as what was shown in the post... I believe I purchased my adapter from this website where it was not clear whether the corresponding IMU is LSM6DSV or LSM6DSV16X...For your convenience, please find attached a photo of my motherboard and adapter.Thanks a lot in advance for your help!

I meet that when follow func is in its second  loop And I find the value of var 'batch' is not 1 in the second loop,Why?  

Hello,I'm considering to use NanoEdgeAI generated model in a Zephyr project but I haven't been able to find any integration with it. I have been trying to integrate it using cmake but I'm having issues to do it (not the same math library, not finding the pointer of the init function, etc). So I wanted to know if is even possible or not what I'm trying to achieve. Regards,Pascal.

Hello, I generated a nClassification model using the LIS2DE12 accelerometer through NanoEdgeAI Studio but when I try to integrate it to my project using VSCode I get the next error:c:/ncs/toolchains/31f4403e35/opt/zephyr-sdk/arm-zephyr-eabi/bin/../lib/gcc/arm-zephyr-eabi/12.2.0/../../../../arm-zephyr-eabi/bin/ld.bfd.exe: ../NEAI_lib/Lib/libneai.a(NanoEdgeAI.o): in function `neai_classification':(.text.neai_classification+0xbc): undefined reference to `fmaxf'c:/ncs/toolchains/31f4403e35/opt/zephyr-sdk/arm-zephyr-eabi/bin/../lib/gcc/arm-zephyr-eabi/12.2.0/../../../../arm-zephyr-eabi/bin/ld.bfd.exe: (.text.neai_classification+0x2aa): undefined reference to `exp'c:/ncs/toolchains/31f4403e35/opt/zephyr-sdk/arm-zephyr-eabi/bin/../lib/gcc/arm-zephyr-eabi/12.2.0/../../../../arm-zephyr-eabi/bin/ld.bfd.exe: (.text.neai_classification+0x2da): undefined reference to `exp'collect2.exe: error: ld returned 1 exit statusninja: build stopped: subcommand failed.As you can see from the log, I'm usin

This is the model i use(through X cube ai 8.1.0)By annalyze it ,the output should be float as belowBut the auto generated code declare it as ai_i8,why?and I can`t get result in the following way,I don`t konw the reason Because of the limits ,I can only provide three files here.I`d appreciate it if someone could answer my questions!!!

Hi all,We have requested FP-AI-FACEREG through https://www.st.com/en/embedded-software/fp-ai-facerec.html with our company e-mail address, but it is not sent although it's been more than a week.Is this a normal process? Is there anyone having the same problem?Cheers,Ogulcan Ariyurek 

i get two defferent results with O0 and O2 optimizationO0 result O2 resultBut I think the result of O2 is more like right because the label is daisy.Should I always open O2 optimization? 

Hi! What is the difference between the part numbers LSM6DSO16ISTR and LSM6DSO16ISNTR? I'm guessing that the one with "N" is able to use the libraries generated by NanoEdge AI Studio, but I can't find a way to confirm this. If that's the case, why can't I see LSM6DSOIS in NanoEdge AI Studio on "Target selection" in "Project settings"? If I can somehow use NanoEdge generated libraries on LS6DSO16IS, can I embed an n-class classification model in it? I'm asking this because I've bought an ISM330IS and then found out that I can only embed anomaly detection models in it. Thank you.

I encountered an error while analyzing an ONNX model quantized with int8, and I'm unsure how to resolve it. Is there any technical support available to assist me with this issue?

I've got a simple model like this in ONNX: I set HCLK@216MHz and ran "Inference on target" using Nucleo-f767zi board with X-Cube-AI 8.1.0I got these useful information:1) duration : 0.017ms (17us) by sample (200 samples, I got a similar result (21us) by a single sample validation)2) cycles/MACC : 10.41 However, when I generated the code, edited the code likeI built the project using X-Cube IDE 1.13.2. I got this:duration: 8us / sample （prescaler set to 216）Why is it significantly faster than that from the default "validate on target" button?1) Why, in the same system configuration in the .ioc file, I got as much as x10 latency using this method?2) Why, in contrast to the average 6-Cycle/MACC for Cortex-M7 in the manual handling float32 data, I got >10Cycle/MACC using Nucleo-F767zi@216MHz?3) Why even a single dense layer using "validate on target" could generate a latency mounted to 8us, a large enough number that could match the mentioned bulit C project? Is

I am trying to use Unico 9.14.6.0 to get the evaluation module to work for the LSM6DSV16X.But I am getting an error that says it's an unexpected device based on the WHO_AM_I register.STEVAL-MKI109V3 updated to v3.48.3STEVAL-MKI227KA (Labeled STEVAL-MKI227AA on board)I have switched the jumpers to the SDx SCx but that didn't seem to work either. Tried both mac and windows.Any help would be appreciated

Hi guys,These days I already know how to generate the AI model named libneai.a(also including the NanoEdgeAI.h) by NanoEdge AI Studio.But I don't know how to use it, I try to import it to MDK, but MDK always has a compiler error. Is there some information about "How to do an AI demo from scratch"?And one more thing, Can I valide the AI model(libneai.a) created by NanoEdgeAI at windows or linux platform?Thanks!

Hello,I´m evaluating multiple accelerometers through the STEVAL-MKI109V3 but unfortunately this is not listed as part of the loggers supported by NanoEdgeAI Studio, this is a bit annoying and I encourage you guys to consider add it. I know this is supported by Unico-GUI and that you can log the files through there but is annoying to adapt these logs every time in order to comply the way NanoEdgeAI Studio process the data.Regards,Pascal.

Hello,I'm working on a project that requires collecting data from the LSM6DSOTR sensor. I want to use the ML capabilities of the IMU, but it seems like I need an ST evaluation board to use the software? I currently have the the LSM6DSO connected to an STM32 board, already programmed with data acquisition. It seemed that I could use NRF Connect to acquire the data, and Weka to obtain the .ufc, but I don't know if this is accurate. TLDR: Do I need an ST evaluation board to use the ML capabilities of the LSM6DSOTR?

Hi, Our team was working on training Deep Learning models for a specific business requirement. We ran into an issue with our training data as it's biased toward our region (as we collected local data). Does ST offer Data Collection services that can help us obtain the required data? We don't want to go with third party data providers (unless they're partnered with ST).

Hello, I'm Alessandro and I'm developing a new application based on LIS2DUX12.I'm using this hw tools: STEVAL-MKI238AA, X-NUCLEO-IKS01A3 + NUCLEO-F401RESW: Unicleo-GUI, UnicoI would ask you if it is possible to decide for which output value of MLC1 decision tree (I've only one MCL tree) it is possible set interrupt INT1.Currently I have an interrupt every time the MLC1_SRC register changes value. I would like to activate the interrupt only for determined output value of MLC1 decision tree. Thanks in advanceAlessandro 

Hello, Hello,I plan to use the VL53L5CX ToF sensor in the machine learning core in lsm6dsv16x. is it possible to use the VL53L5CX  as en external sensor instead of magnetometer. Besides, may I use two external sensor in machine learning core( one will be ToF sensor, another is ST pressure sensor) Regards,Fehmi