Edge AI

Hello,Is there a way to give X-CUBE-AI a model with quantized-weights? Right now the tool is able to do a custom post-training quantization for a given Keras model. However, what if we have a model that already has quantized weights (which is the case when we do a quantization-aware training with Tensorflow Lite for example)? Is there any solution/workaround to make use of "quantization-aware weights"?Thank you and kind regards

Hello,I have noticed that the latest version of X-CUBE-AI (4.0.0) has support for Tensorflow Lite. Therefore I tried to analyze a SSD Mobilenet v1 network that I found on this blog post:Link to blog postLink to .tflite modelWhen analyzing the .tflite model, I have an error: "TOOL ERROR: Unkown layer type CUSTOM, stopping". I suspect that this comes from the post-processing operations that implement the non-maximum suppression.Where can I see the list of supported TFLite operations? Also the list of supported Keras layers were removed from the Getting Started with X-CUBE-AI .pdf.I couldn't find those lists elsewhere. I would really appreciate some help! Thank you and kind regards

Hello, I have some questions about using the code generator I am currently using an STM32F446 and I want to use a Keras model driven by python to make a prediction on the data retrieved by the micro-controller in order to get better results. I found on internet that I should use X-CUBE-AI version 3.4 with ApplicationTemplate as a resource. Is it really the right resource, if I want to do only the prediction on the micro-controller ?If yes, I have, once the code is generated, an MX_X_X_CUBE_AI_Process function that seems empty, do I have to complete it or does it work as it ? If I have to complete it, is it possible to indicate me how to complete it and especially with what.void MX_X_CUBE_AI_Init(void) { MX_UARTx_Init(); /* USER CODE BEGIN 0 */ /* USER CODE END 0 */ }   void MX_X_CUBE_AI_Process(void) { /* USER CODE BEGIN 1 */ /* USER CODE END 1 */ }Where is called the function to retrieve the data from the micro-controller and transform it ? With this, the data can be used by

How Cube-AI is different from ARM CMSIS-NN and ARM NN SDK?

Dear Sirs,I would like to know how to Sci-kit learn training model is handled by x-cube-AI.Thank you for your support in advance !Best regards,Kazu

Hello,I flashed my STM32F4-Discovery. If I want to validate on target I get the following error:"ERROR: no connected board(s) or invalid firmware or the board should be re-started"I think my error comes up with the firmware, but how can I solve it?Best regardsTim

Do we have any list of STM32 devices which which are supported for CubeAI IDE. My requirement is very basic Neural Net to run on device with couple of I2C interfaces and GPIO s. If there is a list of devices with a comparison, it will help in chosing right one for us. Any pointers please?Thank you

Hello,I am trying to install X-Cube-AI on STM32CubeMX.but I am facing this problem when I try to install it from downloaded file, my user account have full access on PC.When I try to install it direct web using CUBMX ,connection is interrupted due to time out.How can I install it?Best RegardsHossam Alzomor

Hello,On ST youtube channel there's a demo for character recognition, image classificationhttps://youtu.be/6uRXvZSU0tw?t=396Is there any resources about this applications?

what does the amount of RAM required depend on?And how it get calculated?

..

Hi, I have installed the latest STM32IDE (V1.0.1) but I cant see how to add the x-cube-ai package. The suggested method is to use the "Help Menu" -> Manage Embedded software packages" but this is greyed out. I also tried using Help->Install new software, I downloaded the "en.x-cube-ai.zip" package and tried to specify it as alocal install, bu this gives errors too.

I've a problem with the generation of the code for the application template. Following the steps, i've generate a blank code, but there's a weird problem, maybe something about linker.In fact looking in Middlewares/ST/AI/AI/src/network.c there are a lot of syntax errors, like if the variables defined in "network.h" wouldn't be considered, even with #include "network.h".besides, calling a function that is defined in network.h, like ai_network_create for example, i get this error "Undefined reference to ai_network_create".With validation template i haven't any problems, i can load the network and validate it both from pc that on the board.here the blank project:https://drive.google.com/open?id=1uMNyChclq7Rt7NMQVxcdyMxSpfZNgoHNhere some errors in network.c

hiMy STM32cubeMX version is 5.2.0X-CUBE-AI.3.4.0But when I follow the steps, I find the information as follows:So what should I do to make it OK?Br,Max.Su

Hi,I am trying to use the Machine Learning Core in the LSM6DSOX sensor. I collected some test data, computed the features using the Unico GUI, but the zero-crossing values don't seem right. I did my own computations with MATLAB and all the other features match nicely.I attached a screenshot of the problem, including a graph of a segment of my acceleration data. Unico says there are 24 zero-crossings in this segment, when there are clearly only 10. Am I misinterpreting this feature, or is it possible that there is a bug in Unico?Furthermore, my data is 253 samples long and the window size is set to 52 samples, so the data can be divided into 4 whole segments. However, the ARFF output file generated by Unico contains only 3. Why is the last segment omitted?Thanks,Balázs

Hi, I am using a Nucleo-F722ZE board and trying two different applications of AI generated by STMCubeMX.So, I have made a simple validation project using a CNN model which was generated, built, and flashed correctly to the board. However, when I tried to run the validation on target from CubeMX it gave an error message "No connected board(s) or invalid firmware or the board should be re-started"Then, I opened the host serial terminal console to check the validation message, it did not print anything. Same happened with the system-performance project as well.I have done those steps for debugging:1. Trying different firmware and AI package versions.2. Checked that the ST-link is working and the board is connected.3. Checked that the firmware is working using different example GPIO, I2C..etc projects.Kindly, find the AI projects in that Github repo. https://github.com/AlyMetwaly/Embedded_AIN.B: I have used the same CNN Model and generated projects for Nucleo-F401RE and it worked perf

I have made a model for imdb dataset in keras that looks like:model = Sequential() model.add(Embedding(known_words_num, word_vector_dim)) model.add(GRU(word_vector_dim, dropout=0.2)) model.add(Dense(1)) model.add(Activation('sigmoid')) model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])Which Embedding layer is used for making word vectors.:|

Hello,I am using CubeMX 5.1.0 with the X-CUBE-AI 3.4 package. It works really fine with Atollic TrueStudio and the GCC. However, the perfomance / runtime of our algorithm is not satisfactory, so that we would like to use the ARM Compiler 6.12 to look if the duration of the algorithm would be faster. Unfortunately, I got some Linker-Errors when I try to use the Keil MDK with ARM Compiler 6.12 and I saw that the Lib-File "network_runtime.lib" is pre-compiled with ARM Compiler 5.06.Is there another file "network_runtime.lib" for ARM Compiler 6.12? If no, when will CubeMX and X-CUBE-AI support the ARM Compiler 6.12?Thanks.Best Regards,Simon

Hello community,I created a Makefile with CubeMX and the X-CUBE-AI software. When I try to compile the project with arm-none-eabi-gcc the linker cannot find network_runtime.a./usr/lib/gcc/arm-none-eabi/5.4.1/../../../arm-none-eabi/bin/ld: cannot find -lnetwork_runtime.a collect2: error: ld returned 1 exit statusThe file network_runtime.a is in the folder Middlewares\ST\AI\AI\lib and the option -L/mnt/DiscoveryBoard/Middlewares/ST/AI/AI/lib/ (absolute path.., I also tried others...) is defined.Maybe somebody knows what I'm missing.Thank you very much.Oliver

Hello. I will be attending the electronic fairs in April in HK and noted that ST has a booth for AI (SMART Living). Are there presentations that we can attend for AI? If yes, what is the schedule? Otherwise, will drop by for a quick chat. Thanks and see you then!

I generated my code (HAR model) and could validate the process with random numbers. But, when I want to validate with my custom data, there is an ''VALIDATION FAILED'' error. please find the attachment contain the error and my csv test file. The file contain 3-axis of data for jogging and contain 62 rows of samples. please let me know what the problem is.Also I will appreciate if you upload a test csv file for custom data validation of HAR for better understanding. Thank you in advance

I want to create an application to train the neural network based on environmental sound.Can you please give me a higher level idea how to train the neural network on any platform and generate the model.

Hi,

I am trying to deploy a pre-trained Mobilenet model using STMCubeMX.ai, but I am facing some issue that I want to report.Here the link of the projecthttps://github.com/alessandrocapotondi/mobilenet_stmcube_aiHere the model I am using:https://github.com/alessandrocapotondi/mobilenet_stmcube_ai/blob/master/keras_model/mobilenet_160_0_25.h5Here the resource usage of my model:600KB RAM for the activations and 1.79MB of weights. Unfortunally despite the "Green Light" given by the tool for the deployment on the STMH7432ZI on the device there is no segment that can fit such ammont of contigous data./* Specify the memory areas */ MEMORY { DTCMRAM (xrw) : ORIGIN = 0x20000000, LENGTH = 128K RAM_D1 (xrw) : ORIGIN = 0x24000000, LENGTH = 512K RAM_D2 (xrw) : ORIGIN = 0x30000000, LENGTH = 288K RAM_D3 (xrw) : ORIGIN = 0x38000000, LENGTH = 64K ITCMRAM (xrw) : ORIGIN = 0x00000000, LENGTH = 64K FLASH (rx) : ORIGIN = 0x8000000, LENGTH = 2048K }Best,Alessandro