Edge AI

I have a STEVAL-PROTEUS device that I collected data with and NanoEdgeAI returned a model that has very high accuracy (almost 100%). In fact, during the validation step on NEAI, the emulator returns very accurate results as well. However, in the deployment step, after compilation (I left everything as default but these are the settings), and replacing the the NCC stub library provided in FP-AI-PDMWBSOC, the results are very poor and totally unlike the results from the emulator.Has anyone seen this or has any idea what might be wrong?

Hi everyone,I am using version 4.5.0 of Nano Edge AI with the STEVAL-PROTEUS1 sensor kit. I am trying to implement anomaly detection for motor vibration. However, when I attempt to add a signal using USB (serial), I occasionally receive meaningless 100-line samples, and sometimes I do not receive any samples at all. Does anyone have an idea of what might be causing this issue? 

In CubeMX I have installed the AI package. After that I go to "Start My project from MCU" then I go to "Middleware" then "Artificial Intelligence" and Enable it.Here, I select whatever model and click Analyze. The button becomes grey, but I don't get any error or other message. According to the videos I saw online, it should analyze the model and compress it.  For the pre-trained model, I tried my own models and the models from the "stm32ai-model zoo" git repository. The Runtime I want to use is "STM32Cube.AI MCU". I have tried all the combinations, and all the models in the repository have the same results. stm32ai-model zoo = \stm32ai-modelzoo\object_detection\pretrained_models\   

I have a large PyTorch model that I pruned using unstructured methods to decrease its size. The model size can only decrease if I filter out the zeros and save the model in a sparse format.However, my problem is that ONNX doesn't support the sparse format. Does anyone know how I could prune a model, filter out the zeros to reduce the model size, and implement it on an evaluation board?  

How to input validation image dataset in the stm32cubeMX.    

When I try to launch the "train.py" file, in conda virtual environment it gives me an error "ModuleNotFoundError: No module named 'logs_utils'". This is a module called on the code line 23 of the train.py. After that I've proceed with a "pip list" to see the modules instaled, but indeed the 'logs_utils'" isn't there.I've tried to install it with "pip install logs_utils" but it simply can't find the module.I've google it and the only module that it seems to exist is the "log_utils".Any one knows how to turn around this?Thank you. 

Hi,Is it possible to add the custom post-processing function in the middleware of the object detection application https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/stm32ai_application_code/object_detection?Thanks! Reply  

It's a shame that the serial emulator in AI Studio only supports a COM port as input and not an offline .csv file of recorded data, as it's not always possible to have access to a live stream.I've noticed as well that the COM port options don't seem to include COM ports that have redirected from a host computer when using Remote Desktop (but TeraTerm appears to do this OK).

Why does the datalogger crash when I put a device_config.json file on the SD card ?I created the device_config.json file using the Python GUI tool, saved it to the computer and then copied it to the root folder of the SD card. The only changes I made was to the accelerometers where I changed the range to 4g (with the exception of IIS2ICLX which was left at 3g).I put the SD card in the STWINBX1 and pressed the RESET button.When I try to start the log (by pressing the USER button) the software seems to crash.The green LED (LED_1) stays in the state when the USER button was pressed (i.e. permanently on or off).The board has to be reset, but then the same happens again if I try to start a log.When I try to access the log files on the SD card on the computer, I get an error saying the directory is corrupt, so this probably shows that the software is crashing...I cannot delete the folders, so I have to reformat the SD card.I have attached a copy of the device_config.json

Hi,I am trying to deploy a pytorch deep learning to stm32. I first converted it to an onnx model and after that verified it in STM32Cube.AI Developer Cloud and in the middle step of optimize it reported the following error. stm32ai analyze --model trans_model_8.onnx --allocate-inputs --allocate-outputs --compression none --optimization balanced --target stm32f4 --name network --workspace workspace --output output STEdgeAI Core v9.0.0-19802 INTERNAL ERROR: Mismatch in input shape of gemm: (BATCH: 1, CH : 12, H: 8 ) x (BATCH: 1, CH: 8, H: 12) When I look at the model visualization in netron I see that the gemm operation is only present in the last linear layer, but that operation is converting 1x702 data to 1x2. I don't know if there's something I'm missing.Would be great if you could assist me in resolving this issue. Thank you!

Hi,I am a research intern working on edge AI for anomaly detection.I am trying to integrate my anomaly detection NEAI model with a Network Management Card project which is in C++(cpp) running on STM32MP1.I would like to know the possibility of integration NEAI anomaly detection static library with a C++ project running on ST MCU.  Thanks in advance!

I am validating a classification model on STM32H745-DISC0 board. The validation is successful but id does not show any accuracy related parameter.It shows all the parameters except accuracy. rmseMaeL2rmeanstd    Ritesh

When converting a generated validation report in Nanoedge ai studio (v4.5.0) I've noticed that the format isn't good because it chops off the right-hand side of the confusion matrix. I can see there's a horizontal scrollbar but it doesn't work in the PDF and resizing the document doesn't show the missing part of the table.I don't see a workaround for this because you can't print the report from the popup window where it does display nicely, all you can do is convert to PDF.I've attached an example of what I mean with the produced report...

Hello, I am trying to train your models on a custom dataset created with​ the STM32F401 and X-NUCLEO 53L5A1 expansion board, and I am new to these things. I have downloaded the hand_posture folder on the model zoo section on GitHub, and followed all the instructions to modify the yaml file and preparing the dataset. But when I run the stm32ai_main.py file it tells me that not all dependencies have been iinstalled, even for other .py files that run with that main file. After I tried to find myself some of these missing dependencies (like Keras, Hydra and other stuff) there are many more missing and I could not find any of these online. Is there a repository to find the necessary files to run the training without errors?

Is there a datalogger using the internal adc of stm32h747i-disco?

Hello.I'm working on the stm32h753vit6 MCU trying to integrate the X-CUBE-AI software package into my project and I noticed a couple of issues.1) After installing the package, I go into the IOC and as soon as I enable the X-CUBE-AI, the RCC_OSC_OUT pin reverts to the reset_state for some reason.2) Also as soon as I enable the X-CUBE-AI, Changes start happening in my clock tree.If I set back the clock tree to the configuration it was before enabling the X-CUBE-AI, My code crashes at the SystemClock_Config method (The method is autogenerated and I don't make any changes to it).Has anyone come across a similar issue?Thank you in advance, Yonatan.

Goodmorning, I have a problem with the validation on target of a CNN network on XcubeAI.I tried to validate the networks on desktop, and it runs, but, when I try to validate on target it gives me that error, I ask you if someone has a solution or something else to try to fix this problem.I'm using version 1.15.1 of STMCUBEIDE and v. 9.0.0 of cube AIthe board I try to validate is UFBGA144. Building project Generating project Temporary project loaded Temporary AI configuration done Generating temporary project code Building project C:/ST/STM32CubeIDE_1.15.1/STM32CubeIDE/plugins/com.st.stm32cube.ide.mcu.externaltools.gnu-tools-for-stm32.12.3.rel1.win32_1.0.100.202403111256/tools/bin/arm-none-eabi-gcc -c -mcpu=cortex-m4 -mfpu=fpv4-sp-d16 -mfloat-abi=hard  -DDEBUG -DUSE_HAL_DRIVER -DSTM32L4R9xx  -IDrivers/CMSIS/Device/ST/STM32L4xx/Include -IDr

I am using the STEVAL-STWINBX1 development board with the FP-SNS-DATALOG2 software V2.2.0.When I try to connect using the Python GUI tool, it gets stuck at "Device Template Model Loading"....I reverted back to V2.1.1 and the GUI tool works.Is there a bug in this new release V2.2.0 ? 

Hi,I am trying to analyse a pre-trained model (Keras, .h5) using STM32CubeMX X-CUBE-AI software and encountered the error “[AI:network_1]  NOT IMPLEMENTED: Unsupported layer type: SlicingOpLambda”. The model is trained with the stm32ai-modelzoo library under object_detection. Attached is the screenshot of the error message.  Any help would be appreciated.Following are some information regarding the versions of the software that I am using.  Python version: 3.10.0STM32CubeMX version: 6.12.0X-CUBE-AI version: 9.0.0Github link for model zoo training steps: https://github.com/STMicroelectronics/stm32ai-modelzoo/blob/main/object_detection/src/training/README.mdtraining_config.yaml settings: 

hello everyone.I created a project for the STM32H747I-DISCO chip, the goal was to run an image classification model on it.I used a model from model zoo. and did quantization on it following the instructions there.The part where I am stuck is when I generated the code through CubeMX then opened the project in CubeIDE and build the project (for CM7 core) but how do you flash the chip? I tried running the project then disconnected the chip and plugged it again, but nothing happened (and nothing on the screen) :thinking_face:.

Hello! I recently used stm model zoo and deployed an image classification model on a chip. The flashing part is made automatically if the chip is connected, but I was wondering how to do it manually (using another software)?I get these folders after processing the deployment code but I do not really understand which files to use. If someone familiar with Model zoo can help me figure this out!Thanks in advance!

I want to use X CUBE AI VERSION 7.3.0. I have installed the package with correct version but when I try to use it pops a saying "Please select local zip file for X-Cube-AI stm32ai utilities". I cannot find the zip file anywhere where should be located. I have also tried version 9.0.0 and no tab opens up in that version. But I need to use 7.3.0. I am attaching a screenshot as well. 

Hello, im trying the examples from Fp-ai-facerec but it only have binaries files, so i could'nt understant how it work to config. PLease if anyone have code of this packs please dm me, thanks a lot :3

I am currently having an issue with flashing my neural network (saved as a .tflite model) to my Nucleo-L476RG board. I am able to analyze the model successfully, but when I select, "Validate on Target", I receive the error message below: E200(ValidationError): stm32: Unable to bind the STM AI runtime with "network" c-model: []  connection to "serial:115200"/"115200" run-time fails  E801(HwIOError): Invalid firmware - COM2:115200 I have the following screenshots to show my STMCubeIDE as I attempt to validate on target: If someone knows how to fix this issue, please let me know. 

Hello, Iam Sudarshan, A working professional and have recently got interested in TinyML. So me being curious, i wanted to interface camera and LCD with my STM32F411RE Nucleo board to run AI models on it.As per my understanding i would be able to interface STM32 camera module and an LCD using I2C or SPI protocols. Can you please tell me if i'am heading towards right direction?