Edge AI

Good morning everyone, I am trying to follow the example for machine learning given for the SensorTile Boxhttps://github.com/STMicroelectronics/STMems_Machine_Learning_Core/tree/master/configuration_examples/example_2_sensortilebox_algobuilder_unicowith Algobuilder.I have followed all the steps with the configuration that did not give problems but when I go to compile the project it gives me this sequence of errors and I cannot flash the SensorTile Box.Firmware Directory Initialization:Firmware Directories Successfully Initialized!Code generation finished successfully.Build Process Started ...Create.Opening 'SensorTile.box-Project'.11:51:40 **** Build of configuration SensorTile.box-Project for project SensorTile.box-Project ****C:/Users/fr.silvano/STMicroelectronics/AlgoBuilder/FirmwareTemplate/Middlewares/ST/STM32_USB_Device_Library/Core/Src/usbd_core.cC:/Users/fr.silvano/STMicroelectronics/AlgoBuilder/FirmwareTemplate/Middlewares/ST/STM32_USB_Device_Library/Core/Src/usbd_ctlreq.cC:/

so, i have been using nano edge ai since a long time. but now i cannot log the signals when i am trying to. will they be any reason for that. please, solve this.

DearI have encountered some difficulties while using CUBEAI to analyze my self-trained object detection models.I trained my own dataset using the object detection models available in the STM32AIModelZoo. I have utilized the tiny_yolo, st_ssd_mobilenet, and ssd_mobilenet_v2 models, and successfully trained the corresponding models in .h5 format. The entire training process should be error-free, as I strictly followed the configuration files provided in the modelzoo's guidelines.However, when I attempted to analyze the trained models in CUBEAI, it displayed an error stating that Lambda layers cannot be processed by CUBEAI. This error persists regardless of the model I use. After reviewing the .py files of various models, I did not find any functions related to creating Lambda layers. I also tried analyzing the pre-trained models provided in Modelzoo with CUBEAI, and discovered that they also cannot be processed by CUBEAI due to the presence of Lambda layers.This issue has been plaguing m

I am currently working on a project using the BL4S5IIOT01A board, where I need to send data, which is being printed on the console, to AWS IoT Core. For this purpose, I intend to utilize the X-CUBE-AWS package.However, I am unable to find the X-CUBE-AWS package within STM32CubeIDE. Could you please guide me on how to integrate or access the X-CUBE-AWS package in STM32CubeIDE for the BL4S5IIOT01A board? Any detailed instructions or steps to resolve this issue would be highly appreciated. i have a normal embedded led blinking code i have attached the document for your reference, please provide a guide how to link this project with aws. i refered https://catalog.us-east-1.prod.workshops.aws/workshops/b5857416-f2f6-4188-8562-053ec5f983ef/en-US workshop but they did not show the clear step like how to link data logger code with x cube aws package. it would be great if you provide assistance on this query 

I would like to use LSM6DSOX with a linux board or potentially another micro controller.  What problems would i run into that using a stm32 couldn't have. I see there's a lot of code generation for stm32 chips. 

Hi there,I have a work-well project with TouchGFX and FreeRTOS.After I enable the CUBE-AI software pack(can't check the application), and compiler the project, the error as below(there are 2 same RW_RAM):After commenting one RW_RAM and compiler again, the error was as follows:I don't know what happened and how to resolve it, can someone help me?Thanks for your help!  

Hi, we are studying option of upgrading one of our existing projects that already incorporates the VL53L5 sensor with the STs EdgeAI machine learning library, that would enable our product to use data from the ToF sensor to learn patterns and detect and recognise various objects, placed in front of sensor. Is this feasible with the NanoEdgeAI studio? In ideal world, we would use full 64 pixel resolution of sensor, but that might be too much of a load on MCU, so at least 4 or 6 pixels would probably do it.

I'm working on CBM model for predicting maintenance on Motor and trying to analyze the bearing frequencies I have few questions to develop a solution based on STWIn-BX1 1) Mounting of the Board - is it necessary to mount the board on the Motor near the bearing? 2) the operating temperature range of the board is just up to 35degc, but the accelerometer and vibratory sensor is rated for 105degC, how to use the board for extended temperature range? 3) I currently use FP-data log firmware to capture the data and do analysis with math lab to predict the bearing frequencies. is there any way i can create a model to include the list of bearing frequencies that will enable me to create failure alerts based on magnitudes related to bearing and implement a decision tree. Regards Dinesh Kumar 

I came across a weird issue when using cube-x-ai. When function ai_network_run() was called, almost everything except ai_output[] in the RAM become messed up, including global variables, static and extern variables. All the peripherals need to be re-initialized because values in the handles are wrong. I don't know what's happening inside the ai_network_run() function because its definition is in "NetworkRuntime900_CM7_Keil.lib". My toolchain  is CubeMX 6.11 + cube-x-ai 9.0.0 + Keil MDK. I have enabled MPU and CRC for CM7 and the configuration is below. Can anyone tell me what is the exact error? Thank you very much!    

I am having a terrible time analyzing a model using Cube AI. When I analyze the TFlite model, I get an internal error. When I analyze the same model saved in Keras, I get the error below. However, the model trains and works perfectly for inference in tensorflow.I have decorated the classes, for example, this way:@tf.keras.utils.register_keras_serializable(package="Custom", name="PatchEmbedding"), but it still doesn't work.Can anyone help me out?

Hi. Can I use the FP-AI-VISION1 package with the STM32F746 board? If not, how can I make a computer vision solution with the STM32F746 board?

Hi, I am loading a .ucf file on the STWIN using the ST BLE Sensor Classic app, and then performing data logging using an SD card. In the resulting SD card files (STWIN_XXXXX) I get some files, including ISM330DHCX_MLC.dat, the outputs/classifications of the MLC that I want to save (instead of just viewing them in the app).  Using the HSDPython_SDK examples from the FP_SNS_DATALOG1 package, I am able to export the .dat files for the raw sensor data (e.g. ISM330DHCX_ACC.dat), using the "hsdatalog_data_export.py" script, but not for the ISM330DHCX_MLC.dat. file (which I can plot and view as a dataframe, but I am unable to export it to a .csv/.tsv file). Can you let me know how I can export the contents of the ISM330DHCX_MLC.dat file to a .csv/.tsv file? Is there a specific command line argument I should specify, or should I tweak something inside the "hsdatalog_data_export.py" script? Thanks in advance.

I have a DB (Differentiable Binarization) model exported to ONNX, and I quantized it with onnxruntime using the code below:```dr = DetDataReader(calibration_dataset_path, input_model_path) # Fake images as I'm just attempting to measure the memory usage.conf = StaticQuantConfig(calibration_data_reader=dr,quant_format=QuantFormat.QDQ,calibrate_method=CalibrationMethod.MinMax,# optimize_model=True,activation_type=QuantType.QInt8,weight_type=QuantType.QInt8,# nodes_to_exclude=['resnetv17_dense0_fwd', ..],# nodes_to_quantize=['resnetv17_dense0_fwd', ..],per_channel=False) # I tried True as well. quantize(input_model_path, output_model_path, conf)``` Then I tried to analyze it with the stm32ai tool, but I got the following error$ stm32ai \analyze --model model-quant.onnx \--type onnx \--name det \--compression high \--no-onnx-optimizer \--output output \--allocate-inputs --allocate-outputs \--with-report \--series stm32f4 --target stm32f4 --workspace workspaceNeural Network T

안녕하세요 제목과 같이 관련돼서 문의 합니다. 입력 데이터는 아래와 같고,Translated to EN: Hello I have an enquiry related to the title. The input data is as below, // 13 x 15 int new_data[195] = { /* fill with your sensor data */440, 501, 743, 479, 546, 658, 596, 523, 493, 684, 855, 3, 2,536, 605, 743, 576, 664, 791, 728, 523, 591, 799, 993, 4, 2,536, 605, 743, 576, 664, 791, 728, 523, 591, 799, 993, 4, 2,659, 735, 924, 707, 664, 965, 728, 651, 722, 961, 1184, 4, 2,659, 735, 924, 707, 827, 965, 901, 651, 722, 961, 1184, 4, 2,826, 735, 1168, 887, 827, 965, 901, 827, 901, 961, 1446, 4, 2,826, 920, 1168, 887, 1059, 1198, 1128, 827, 901, 1192, 1446, 5, 2,826, 920, 1480, 887, 1059, 1198, 1128, 1084, 901, 1192, 1736, 5, 2,1067, 1174, 1480, 1130, 1395, 1507, 1407, 1084, 1151, 1519, 1736, 4, 2,1067, 1174, 1480, 1130, 1395, 1507, 1407, 1084, 1151, 1519, 1736, 4, 2,1413, 1510, 1765, 1443, 1395, 1790, 1663, 1437, 1480, 1778, 1950, 4, 2,1413, 1510, 1765, 1443, 1686, 1790, 1663, 1437, 1480, 1778, 1950, 4, 2,1711, 1510, 1996, 173

Hello everyone,I'm trying to use the sensortile box pro for an ML alghoritm.My idea is to have a machine learning algorithm that can identify two types of vibrations: correct and fault.Using the firmware loaded on the Sensortilebox Pro Datalog 2.1.1, I was able to obtain the values from the accelerometer and gyroscope lsm6dsv16x_gyro.csv and lsm6dsv16x_acc.csv along with those from all other sensors such as the magnetometer lis2mdl_mag.csv.Using a Python script, I then verified that the timestamps for the accelerometer and gyroscope were the same and merged the CSV files but i cannot do the same with the magnetometer. To create a decision tree, I have split the merged CSV into two different CSV files, one for each type of vibration: lsm6dsv16x_all_correct and lsm6dsv16x_all_fault.Using the Unico_GUI software, I would like to generate the UCF file needed by the Sensortile to distinguish these vibrations. I followed the guide on GitHub for machine learning: https://github.com/STMicr

..

I am facing some issues in my first AI project implementation using UM2526Starting a new STM32 AI project, can you anyone suggest the resources to get started

So i have a LSM6DSOX and connected to its sensor hub LIS3MDL. Currently I'm trying to log data to feed into the NanoEdge studio.  So i need to grab data from the actual LIS3MDL. On the sensor hub you can read and write registers through the sensor hub correct?

Is it possible to implement the model of human activity recognition mode zoo on the STM 32f446RE board Only the board and one sensor are used.Is it feasible or not?

I am using STM32CubeIDE v1.12.1 on macOS. I have noticed that, when I import a project generated with STM32Cube.AI Developer Cloud, whenever I open the IOC and make a change to the device configuration that requires to regenerate the code, the project becomes broken. I therefore did an experiment: I created a STM32 project under the IDE, then I enabled the X-CUBE-AI software package in the device configuration tool and I let the IDE generate the code. Apparently the IDE-generated code is missing the $PROJECT/Middlewares/ST/AI subdirectory with the X-CUBE-AI library and headers. I have the same effect with 8.0.0 and 8.0.1 versions of X-CUBE-AI. How do I work around this issue?Best,Pietro

good morning, I'm trying to import a neural network for NILM (non-intrusive load monitoring) in the stm32 environment, but I can't complete the analysis due to the presence of the error: INTERNAL ERROR: unpack_from requires a buffer of at least 122836864 bytes for unpacking 4 bytes at offset 122836860 (actual buffer size is 1654784) The network is CNN type and I'm working with xcube ai v. 9.0.0 and stm32cubeide 1.15.1I'm asking anyone if they know how I can solve the problem of the buffer size, thanks in advance.

Hello,I am trying to apply a CNN+GRU network based pre-trained model (Keras, .h5) onto the board (STM32G4474RETx) by using X-CUBE-AI and encountered this error: "Unsupported wrapped layer type: Conv1D" when I tried to analyze it.Based on the error message, I guess the TimeDistributed layer is not supported.Is there any way to make it work? I would greatly appreciate any help. Thank you in advance.

LIS2DUX12 -interrupt used for stepcounter which are the interrupt used for step counter design

hi I want to start with LIS2DUX12 accelerometer programming ,It is not clear for me from where should I start . could anyone please tell me what I should do ? Regards

Board - B-L4S5I-IOT01A BoardAI application tool - NanoEdge AI StudioCUBEMXIDE - STM32CubeIDE 1.15.1I made one model.Model detect anomaly on nano studio.I deploy model. Deployed system always show similarity a fix value. Sensor data change. Detection fail.Sensor store value in int16 type variable. Sensor value is not float. Please tell to correct model.Plse check program on link - https://drive.google.com/file/d/1vsZeAcgctPcmEx4skoKoW3RYAxauyKh1/view?usp=drive_linkPlease see log file teraterm.log