Edge AI

As you see,i used cubemx create project which used cubeai.But i don't konw how to used it.What libraries are included in the code generated by CUBEAI? How should I structure my code and what is the underlying logic? What types of return values should I expect? Are there specific explanations, documentation, papers, or tutorials available for reference?

Hello, I am using X-CUBE-AI to run an AI model on an embedded system. While executing the following line of code: batch = ai_mnetwork_run(net_exec_ctx[0].handle, ai_input, ai_output); a Hard Fault error occurs. What’s puzzling is that the error initially occurred, but without any code modifications or corrective actions, the issue temporarily resolved itself. However, after some time, the same Hard Fault error reappeared. Has anyone encountered a similar issue, or could you suggest possible solutions?I would greatly appreciate your help. Thank you!

When i used cubeai,i met some problem which i click anaylze the error is  "Unsupported layer types: DynamicQuantizeLinear, ConvInteger".And i saw my model's quantizelinear manner is dynamic.so i want used pycharm to solving my error.i want make a static quantization and pruning to my trained YOLO model.my model's type is PT file,can you give me a code to solve my problem and i will very appreciate for your help!

Can I run a Machine Learning model using Tensor flow lite or Edge Impulse on STM32F103C8T6 with 64kb flash.Actually I don't have any choice but to try to run the ML model on Blue pill, because my there is unavailability of other MCU boards at the institute. Any kind of help would be appreciated !

After automatically searching for the best NanoEdge AI library, only the library ID and model name are displayed, and the specific content of the model, such as the neural network structure, cannot be found. I want to know about this aspect.Can you help me solve this problem? Thank you.

Under STM32CubeIDE I generated a neural network form my Nucleo board with X-CUBE-AI 9.0.0 from a quantized TFLite model of a sine function. I followed the same steps as I did about one year ago with X-CUBE-AI 8.1.0 (at the time it worked) but this time, when I perform inference, the network answers with constant output independent on its input. My steps are essentially these:- Create a project for my board (Nucleo F767ZI);- Add the X-CUBE-AI 9.0.0 pack and the TFLite model;- Accept all the optimizations that the IDE proposes;- Activate the CRC;- Generate code;- Add in the main() the code to create the network / set the activations, input and output buffers / run the network; The code is almost identical to the example contained in the X-CUBE-AI documentation.My development platform is a MacBook Pro with M3 chip and Sequoia OS. It is likely that I missed something, since one year passed and I possibly forgot to write down some details of the configuration, but I tried all the variations

I currently have the STM32H7 DISC0 and plan on getting the STM32N6 development board. What I can't seem to figure out is if it is possible to train a custom dataset model on my own in tensorflow then download it onto the STM32 for inference? I was looking at platforms like YOLOv11 and Efficientdet. Also, are there any detailed videos or instructions I can look at? Please advise. Thank you. Marcus

Hi,I am currently working with the nucleo-n657x0-q development board and trying to use the ST neural-ART NPU of the STM32N657X0 MCU. While I know that neural network models can be automatically mapped to the NPU through the STEdgeAI-core, I was wondering if there is any documentation on how I could manually program the NPU? Either through a library/API or through low level functions.Thanks in advance!

I am using the STM32N6570-DK board to run AI model on it. And I have modified the stm32n6.mpool file because the board actually have a 1‑Gbit Octo‑SPI flash memory, which is 128MB instead of 64MB.  But even if I have already adjusted it, the cubemx still shows "Used Flash:3.51MB(.00B over 64.00 MiB xSPI2)". Is there a problem and how I can fix it? By the way, I want to know that where I can get some examples about using AI on the STM32N6570-DK board, and the correct way to modify the files so that I can use the external memory on the board.

Hello STM Community,I have a question regarding my FCNN model. My goal is to feed input data into the model and obtain the corresponding output, but I haven’t been successful so far. Could you please assist me with this? Any guidance would be greatly appreciated!Model Layer Information:Stm32 Anlyze Show Graph: 

Hello, I'm writing my bachelor thesis regarding the deployment of the model inside the STM32, and I just want to use CubeIDE to run my own program. It takes in the image data through the UART, put the data inside the model input buffer and then do the LL_ATON_RT_Main(&NN_Instance_Default); and finally gives out the ourt put.I really did everthing I can, I already wrote the parameters generated by the Cube AI in side the External flash (exactly same place as the cubeAI report, which is 0x70000000 ). The model is from offical website:  mobilenet_v1_0.25_224_fft_int8.tflite .the file is from the Neural ART(cube AI) the origianl name is network_atonbuf.xSPI2.raw, I copypaste it and rename it to .BIN so that I can download it with the cube programmer instead of using the cmd lines or some python files of the Neural ART.After that, I run my program, below is my ST program ,It works without any "wornings or bugs",I can inspect the output by monitoring the float buffer[101]

I encountered two problems while learning the Extrapolate library:One of the issues is that I am not very clear about this extrapolation value, and I don't know if there is a problem understanding it. I have always believed that this extrapolated value is a predicted data value made by AI after reading the raw data of my sensor and predicting the trend of sensor changes. However, while studying, I found that this value seemed to need to be set by myself, which made me very confused.Another question that I derived from the first one is that I am not sure what kind of data should be included in Add Singal. In my personal understanding, if I have two columns of data, is the first column the predicted value I set myself? Is the second column my original sequence?Then I tried to search for information in comments and online, but I found that I still couldn't find more information about the Extrapolate library in this NanoEdge AI studio. If you know, can you help me solve these two confusion

Hi, I'm trying to test a model on the STM32N6 but stedgeai tool fails to analyse the model:ST Edge AI Core v2.0.0-20049TOOL ERROR: list index out of range I'm using X-CUBE-AI 10.0.0, STM32CubeN6 v1.1.0 and STM32CubeIDE 1.18.0.The model I'm trying to use is in attachment.I've also tried with STM32Cube.AI MCU Runtime mode but the error is the same.How can I fix the issue ? Thanks,Alexis Murzeau

"Can Cube.AI deploy Transformer models? To deploy the Transformer model, I removed the LayerNorm layers and replaced them with BatchNorm1d layers while fixing all dynamic dimensions. However, during deployment, I encountered the error: INTERNAL ERROR: 'NoneType' object is not iterable."  

Hi All,I am looking at a similar kind of https://www.aceinna.com/openimu simulator for inertia measurement(accelerator, gyro, magneto meeter) for character recognition.I did search in this forum https://community.st.com/t5/mems-sensors/is-there-any-simulation-software-available-for-imu-sensors-where/m-p/314540#M13158 but I could see any response. Advanced Thanks.

X-CBUE-AI无法导入ONNX模型，最新版的X-CBUE-AI不支持ScatterND 和 Range 这两个层

hi,my question is why i can't analyze my model about onnx and cubemx don't back type about my error.my model about onnx size is nearly 3mb,so the compression which i used is high.but the cubemx still can't analyze.emmm i hope their have a handsome guys can help me.i will very appreciate to you!

After I uploaded my own model on the ST Edge AI Developer Cloud platform, I further quantized and optimized my model. However, when I selected my own model and after a few minutes of analysis and identification, the platform could not identify the model's input, output, and model type, so that I could not proceed to the next step. If possible, please let me know the reason and help me solve this problem. Thank you very much. 

I am using the X-Cube-AI to deploy a neural network onto  a STM32N6570-DK board. I selected the "n6-allmems-O3" option, and after analyzing, it shows two data, labeled as used RAM and used Flash. I am wondering since I have selected the "allmems" option, the RAM usage refers to the usage of the external RAM or the internal RAM, and the Flash usage refers to the internal Flash or the external Flash.

I want to use the STM32F407 to implement the NanoEdge AI Library for Anomaly Detection (AD) tasks.The documentation mentions nanoedge.ai's method for backing up and restoring the library knowledge using the .ld file in a Linux environment. I want to know how to perform this backup operation in Keil.

Hello,I'm working with the STM32N6570-DK board and using X-CUBE-AI to run quantized neural networks converted from ONNX models.I'm trying to better understand which parts of the model are executed in software vs. hardware, specifically on this board, and what determines that split.My Observations:1. ResNet-8 Model (custom quantized)When I validated a quantized ResNet-8 model, the output during target validation showed:Total number of epochs: 16 of which 2 implemented in softwareepoch ID HW/SW/EC Operation (SW only)epoch 15 -SW- ( Softmax )epoch 16 -SW- ( DequantizeLinear )So, almost everything was offloaded to hardware, except for the final softmax and dequantization — which I expected.2. Simple Model Quantized via ST Developer Cloud However, when I tested a very simple model (shown above), quantized through the ST Developer Cloud, I saw this:Total number of epochs: 5 of which 3 implemented in softwareepoch ID HW/SW/EC Operation (SW only)epoch 2 -SW- ( DequantizeLinear )epoch 3 -S

Hello,I have started using the feature importance tool provided in the new update of Nanoedge AI Studio.However, it is unfortunate that we can't keep/ download a report of the results.Is it a feature I am not aware of ? Thanks

After upgrading to version 5 of NanoEdge AI Studio, I’m unable to create new projects, and I also can't open projects originally created in version 4.6. Has anyone experienced similar issues?

I have just started learning X-CUBE-AI and encountered an issue when deploying a large model. Currently, I am trying to deploy a model of approximately 4 Mb on the STM32F4ZGT6, but its internal FLASH is only 1 Mb. To address this, I plan to use an external W25Q128 FLASH chip. However, I couldn't find any relevant tutorials, and after compiling the code, I received the following errors:ResNet.axf: Error: L6406E: No space in execution regions with .ANY selector matching main.o(.bss).ResNet.axf: Error: L6407E: Sections of aggregate size 0x21540 bytes could not fit into .ANY selector(s).

Hi everyone!I'm trying to validate an ONNX model on my STM32H747I-DISCO board, but I'm running into issues.I've successfully generated the project and flashed it onto the board. The validation process does run (using random inputs), but the final metrics I'm getting are all NaN.I'm using the FP32 version of the model (not quantized), and when I run the same validation on desktop, the metrics are almost perfect — so the model itself seems fine.I've tried enabling the --classifier option, but it didn't make any difference. I also attempted using --no-onnx-io-transpose, but got the following error: stedgeai: error: unrecognized arguments: --no-onnx-io-transpose Could this be an issue with my model, the CubeIDE project or am I missing something in the X-CUBE-AI configuration ? I've also noticed that there is a note in the csv file with the inputs:# Note: w/o datanumber of item by sample is exceeded: 4096 > 512Is this issue compromising the classification ?Thanks in advanc