Skip to main content
Graduate II
June 12, 2024
Question

ML implementation in ISPU

  • June 12, 2024
  • 2 replies
  • 1429 views

Can anyone provide a documentation over integration custom ml algorithm in ISPU.

If I have a machine learning model like tflite or onnx how can I deploy it in ISPU

    This topic has been closed for replies.

    2 replies

    Technical Moderator
    June 12, 2024
    Amith_lalAuthor
    Graduate II
    June 12, 2024

    I think in this article the are dealing with the example provided in the X-CUBE ISPU. What I need is to deploy the model that I developed (.tflite or .onnx) to the LSM6DSO16IS. What should be the next step? Is there any document related to that?

     

    Technical Moderator
    June 28, 2024

    Hi @Amith_lal ,

    Thanks for your patience. 

    We now have tools to convert pretrained models to an implementation optimized for ISPU. You have to download ST Edge AI Core from here: https://www.st.com/en/development-tools/stedgeai-core.html (available for Linux and Windows, soon for macOS). If you want a graphical interface, you can also download MEMS-Studio from here: https://www.st.com/en/development-tools/mems-studio.html.

    The ST Edge AI Core package includes some documentation. In addition, you can refer to the documentation and resources on the GitHub repository for ISPU: https://github.com/STMicroelectronics/st-mems-ispu. In particular, you can refer to the template specific for the integration of models converted with ST Edge AI, which includes a README file explaining the steps to perform: https://github.com/STMicroelectronics/ispu-examples/tree/master/ism330is_lsm6dso16is/template_stedgeai. At https://github.com/STMicroelectronics/ispu-examples/tree/master/ism330is_lsm6dso16is/template_stedgeai_validate you can instead find out how to validate your converted model.

    You may be also interested in the ST Edge AI Developer Cloud (https://www.st.com/en/development-tools/stedgeai-dc.html), which allows you to benchmark your models and download the generated code with no installation or additional hardware required.