Edge AI

Hello ST community!Currently i'm working on a project based on the stm32 NUCLEO-N657X0-Q which should be delivered very soon .I followed this article https://community.st.com/t5/stm32-mcus/how-to-build-an-ai-application-from-scratch-on-the-nucleo-n657x0/ta-p/828502 step by step with the same versions of software mentioned but without any results .I tried using X-Cube AI and the new tool Stm32Cube AI studio but i ended up with a lot of confusion and errors .I'm planning to stick for now with x-cube ai since the new tool has a lot of bugs to be fixed .My questions is what are the right versions of the software to use to get a computer vision (Yolo like) to work on my board ?Thnx!

 Hello ST Community,I am currently working with STM32N6 and CubeAI, and I would appreciate your guidance on an issue related to incorrect inference on the device.I am integrating a custom KWS model into the STM32N6-GettingStarted-Audio project, but inference results on the device are incorrect.I have carefully verified the preprocessing and model behavior as described below.--------------------------------------------------Model--------------------------------------------------Architecture: DS-CNN (PyTorch)Input tensor:[B, C, H, W] = [1, 1, 40, 49]H: mel binsW: time frames--------------------------------------------------Verified points--------------------------------------------------1. Log-mel preprocessing alignment- Log-mel spectrogram generated on STM32 was dumped using GDB- Compared with Python implementationResult:Nearly identical (very small

Hello,I am trying to integrate a model generated by STM32Cube AI Studio into an existing STM32N6 DC-DC converter control project, but I am not sure what the correct integration flow isfor a Neural-ART / NPU-based model.I am attaching the following 3 files/projects:1. My existing STM32CubeIDE project2. The folder generated by STM32Cube AI Studio3. The ONNX model used for generationMy setup is:- MCU/board: NUCLEO-N657X0-Q- Existing project: test_BOOSTXL_ACM_AI_0- The existing control loop is already running correctly- ADC + DMA + TIM based DC-DC converter control is implemented- The control calculation is currently executed inside the ADC DMA interrupt handler- I trained a small 1D CNN in PyTorch, exported it to ONNX, and then used STM32Cube AI Studio to generate codeModel information:- Input shape: (1, 1, 64)- Output shape: (1, 2)- Input data: 64 current samples at 5 us interv

Helloi impported a TFLITE image classification model to cubeAI studiobut everytime  i press runi get this error : Run failed: Error: analyze error: TOOL ERROR: C:\Users\gouiaa\Downloads\mobilenetv2_a035_128_int8.tflite does not contain a valid tflite modeldespite selecting a TFLITE model like the caption showshere is my settings windowMX 6.17 IDE 2.1.1 programmer 2.22 edgeAI  core 4.0

I have been playing with the n6-dk for a while and today something unexpected happened.When I try to run the stock code provided in the model zoo services, the board flashes for a moment at startup and the debugger and the board become unresponsive (requiring power cycling).Another weird thing is that if I flash the code and connect the board to a power source the code seems to be running correctly. I tried with another desktop experiencing the same issue. I also tried to do a mass erase and fresh flashing of fsbl, source, and network code.This was not a problem before, for weeks I run the code with the board connected to my desktop, I have never seen the code crash like that (normally, if something in the code is wrong, I hit a fault and the board get stuck in a sig handler loop).Has anyone experienced anything similar?

Hi,I'm currently trying to run an image classification AI on a STM32H747I-DISCO board with the B-CAMS-OMV camera module bundle, using my own dataset. I was told it's good practice to use the same sensor for dataset collection and the final application, so I want to collect the images for the dataset with the B-CAM-OMV camera module aswell.I've seen that the model zoo provides many applications for the DISCO, unfortunately no simple USB-Webcam. I found such an application in the deprecated FP-AI-VISION1 pack, but it seems that my DISCO is a newer model and has another LCD (MB1166-A09), which doesn't work with the FP-AI-VISION1 pack.Does anyone know of a similar up-to-date application for simple taking images or has a different solution?

Is FP32 forcing software kernels and preventing NPU acceleration? Hi,I am testing CIFAR-10 ResNet on STM32N6570-DK and I need help understanding a latency gap between two STEdgeAI generation flows.I used the same FP32 model as input:CPU-only generation flowCPU+NPU generation flowBoth flows generated C files and hex successfully, and both run correctly on target.However, the CPU+NPU build is significantly slower than CPU-only.What I would like ST to confirm:For STM32N6, can full FP32 ResNet graphs be truly accelerated on Neural-ART, or are they generally expected to run as software float kernels?Is it normal that CPU+NPU generation for FP32 can be slower than CPU-only generation?For real NPU acceleration, is calibrated INT8 mandatory in practice?Which exact generator settings s

Hi I am using NUCLEO-N657X0-Q board, for the STM32N657 part, with the release STEdgeAI 2.0.We are using STM32Cube ecosystem (IDE, CubeMX, Programmer) for the development. I have been using NPU Validation example without any issues. The example there for the DK board, "\Projects\STM32N6570-DK\Applications\SNS\", which does not run for NUCLEO after a bit of changes for Nucleo. Gets stuck after the SystemIsolation_Config() in AppliNonSecure, with a Hard Fault. SECURE_RegisterCallback(IAC_ERROR_CB_ID, (void *)SecureError_Callback); (which is IACCVIOL = 1) ==> Though I have MPU is already enabled in the FSBL.    Same time, after a bit of struggle with the CubeMX, Generated the

I have followed the instructions mentioned in :Getting started - How to evaluate a model on STM32N6 development board for a quick profiling of my model.Board: STM32N6 Nucleo boardI, however , get a timeout message "LOAD ERROR: STM32 - read timeout 50047.0ms/50000ms" when running the Ai runner. I have done the following.1. In Dev Mode(BOOT0 and BOOT1 set) , n6_loader.py is executed and the log is attached as n6_loader.log2. The image Post_programming.png shows the output in TerraTerm after loading the Software.3. After setting the relevant python paths, and running the validate command as below, I see the load error. Attached is the screenshot Ai_runner_log.png.$ stedgeai validate -m <model_path>

***Update***This issue is solved and was due to improper wifi firmware upgrade. Topic can be closed.   Hi STM Team, I have been having issues for the last week with connecting to the STWIN.box via wifi from both iphones and android phones for DATALOG2 v3.2.0. (And to some extent v3.1.0). The configuration of the wifi network does not work correctly when I attempt to send it from the app. The only way I could get it to work, is when I used the Python SDK with firmware version 3.1.0. I was able then to connect with FileZilla and view/download files.Please advise what the possible workarounds are with this issue as I need to use this sensor by connecting to it via wifi and downloading the data files via FTP server.The following is an AI-generated overview of the issue in more detail: Device: STEVAL-STWINBX1 (STM32U585AI) Firmw

As shown in the figure, I have equipped the MCU with an external FLASH connected via XSPI2. However, it seems that AI Studio is not aware that there is an external FLASH. How should I configure it to solve this problem?

After I generated a complete project using STM32CUBE AI studio, I noticed that the code for calling the model in the main function was named STM32CubeAI_Studio_AI_Process.After I found its definition, I discovered that it was defined in the file app_x_cube_ai.c. When I examined its implementation, I found that there was no implementation inside the functions acquire_and_process_data() and post_process().Furthermore, I discovered that it did not utilize the so-called latest stai library. Instead, it directly used a more fundamental implementation. However, I am unsure of how to implement the above two functions. I haven't found any relevant examples or tutorials either. Is there any relevant link or document that can help me resolve my confusion?Additionally, I would like to know if it is possible for me to use the STAI library to implement the functions of the aforementioned two procedures.

My development board is STM32MP257F-DK, and I want to use ST Edge AI Developer Cloud to convert. onnx models to. nb models. However, during the interface conversion of 'Optimize your model with STM32AI MPU Tool', an error message 'Error while generating optimized file. Generation does not contain any output' was reported. Afterwards, I will use the model provided by ST Model ZOO: st_yoloxn_d033w_025-416_qdq_int8_ort_detection_CCO_2017_Serson. onnx. I also encountered an error while optimizing. What is the reason behind it? A few days ago, I was able to successfully convert ssd_mobilenetv1_pt_coco_300_qdq_int8_ort_detection_CCO_2 The error message is as follows:

Whenever I am starting to make project  from scratch in STM32Cube AI studio, It gives complete structure with .ioc files as well as AI files, but as soon as I make changes in .ioc file and generate code after loading in CubeIDE, it changes the middlewares and build error comes . What is the solution for that? 1. Should I first make the .ioc project and then make project in STM32Cube AI studio ( Import the existing project from STM32CubeMX or STM32Cube AI studio)2. Another doubt is if I make project from preexisting .ioc, Do I need project in application mode for that as AI changes will work in application mode only ( but STM32N6 doesnt work without FSBL), what is your suggestion for Edge AI project building? The error was :Description Resource Path Location Type make: *** No rule to make target 'C:/Users/hp/.stm32cubeaistudio/workspace/Modified_STM32/Middlewares/ST/AI/Npu/ecloader.c', needed by 'Middlewares/ST

I've implemeted a vision transformer for STM32N6 NPU.Repository: https://github.com/minchoCoin/stm32n6-transformerBatchMatmul operator is not yet supported on NPU and run on the CPU, but it is much faster than STM32H7(inference time was reduced by about 90% than STM32H7) because fully connected layer and convolutions is executed on NPU.Vision Transformer for STM32N6 NPU has three differences from the original ViT(A. Dosovitskiy et al., 2020)1. Patch Embedding (PE) is configured to be performed in the pre-processing process before model input: Because if the model has PE and self attention, it is not possible to interpret the model structure using STEdgeAI (I don't know why yet...)2. Remove bias parameter from Fully Connected layer or change fully connected layer to 1x1 Conv2D:  fully connected lay

Hello ST Community members,   STM32Cube AI Studio is ST’s next-gen desktop solution for optimizing and validating AI models running on STM32 MCUs. It brings a unified workflow, improved performance evaluation, and a modern user experience designed that will benefit both embedded engineers and data scientists. STM32Cube AI Studio is fully compatible with the STM32Cube ecosystem, allowing a smooth transition from your existing projects. Learn more: 

Hi I have FP32 TFLite model. I can able to compile quantised model with STM32 Cube AI studio. i am getting some error while compiling FP32 model with STM32 Cube AI studio. Please let me know the steps to build FP32 model and make the C files correspond to the model.  

Hi everyone,I’m having an issue connecting my custom board based on STM32H723ZGT6 to STM32Cube.AI Studio. The software cannot detect any COM port or target device.Setup:UART interface via USB-TTL converterUsing USART1 (PA9 TX / PA10 RX)TX/RX crossed correctlyWindows Device Manager detects the COM port normallyThe same COM port appears and in other software (e.g. Arduino IDE) 

Board: STM32N6570-DKIDE/Toolchain: Zephyr RTOS + west, Zephyr SDK 0.17.4Repository: stm32-hotspot/zephyr-stm32n6-ai-people-detectionOS: Windows 11I am trying to run the STM32N6 AI people detection application in flash boot mode using MCUboot RAM_LOAD as described in the README. The application runs correctly in debug/development mode (west debug), but does not boot at all when built and flashed for boot-from-flash mode.No output is seen on the UART console after the MCUboot banner — inference never starts.Steps to Reproduce:Following the README exactly:# 1. Build west build -b stm32n6570_dk --sysbuild app

Hello,I have my N657X0H3Q running on Cube AI Studio, the model to use is .onnx file, and the connection to the board is checked.However, there's an error while Cube AI Studio automatically programming the "network_atonbuf.xSPI2.bin" by cube programmer. But I can manually do it by cube programmer without Studio. Following file shows the error message. Thanks for your help.

I'm new to using AI. I want to create a USB adapter that will connect to devices with a USB port and manage and control them.How would I go about accomplishing this?Are there any pre-defined models I could use?

Hello,the board type is STM32N647A0H3Q, and before creating the project it's already turn on the dev mode as shown below.To validate ST link work with my PC, the STM32 Cube Programmer successfully connected to the board.

I’ve generated a project with Cube AI studio for a Nucleo N6 board, but , when I enable SAI2 peripheral in I2S mode, I notice from the oscilloscope that the bit clock line and FS line become corrupted/glitch when using NPU for inference, could somebody explain me why and how to solve it? I’ve used the same SAI configuration for another project without any AI algorithm and the signals from the pins are clean. I don’t know why, it’s a problem that I’ve been stuck for a month or so

Hello Everyone,I’m still getting up to speed with the STM32N6 series, so apologies if my question seems off, but I’d really appreciate your help!I’m working on implementing object tracking using the STM32N6570-DK board with the MB1854B camera. I'm wondering if it’s possible to deploy OpenCV on this platform to support the tracking logic.If anyone has experience with this or knows of any courses, guides, or resources that could help me deploy OpenCV (or similar tracking frameworks) to the board, I’d be incredibly grateful.Also, if you have relevant experience and are open to offering support or consultancy, I’m more than happy to discuss compensation. Please feel free to reach out!Thanks in advance for your time and help!

I've tried several models provided in STM32 ai model zoo. I'm wondering why some models does not show the acceptable accuracy. My test results are as follows.efficientnetv2/Public_pretrainedmodel_public_dataset/ImageNet/efficientnetv2b0_224/efficientnetv2b0_224_qdq_int8.onnx  --> Not Goodmobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenetv2_a100_224/mobilenetv2_a100_224_qdq_int8.onnx -> Goodmobilenetv2_pt/Public_pretrainedmodel_public_dataset/ImageNet/mobilenetv2_a100_pt_224/mobilenetv2_a100_pt_224_qdq_int8.onnx -> Not Goodresnet50v2//Public_pretrainedmodel_public_dataset/ImageNet/resnet50v2_224/resnet50v2_224_qdq_int8.onnx -> Good My testing environment- stm32ai-modelzoo : 4.0- stm32ai-modelzoo-services : 4.0- STEdgeAI : 3.0- STM32CubeIDE : 2.0.0- Windows 11And in order to generate network file and bin file, I used the python file in stm32ai-modelzoo-