This repository contains Machine Learning Evaluation Kit (MLEK) pack which provides CMSIS Reference Applications and templates for Edge AI development.
The provided applications implement data preprocessing, memory management, and neural network inference pipelines that are optimized for Cortex-M and Ethos-U platforms. The examples are prepared to run with Keil Studio.
The CMSIS-MLEK software pack is derived from the Arm® ML embedded evaluation kit and makes the examples easier to access. It also contains interfaces to physical hardware and simplifies porting to target hardware. It contains the following ML applications and uses currently Neural Network Models currently in TensorFlow Lite format.
| ML application | Description | Neural Network Model |
|---|---|---|
| Keyword spotting (KWS) | Recognize the presence of a key word in verbal speech | MicroNet |
| Object detection | Detects and draws face bounding box in a given image | Yolo Fastest |
| Generic inference runner | Code block allowing you to develop your own use case | Your custom model |
Refer to Overview for more details.
| Directory | Description |
|---|---|
| .ci | Files that relate to CI tests. |
| .github/workflows | GitHub Actions for validation and publishing. |
| overview | Top-level overview of MLEK reference applications. |
| docs | Source of the user documentation of the MLEK reference applications. |
| template | MLEK reference applications source code. |
CI Workflow |
Description |
|---|---|
| mkdocs | Builds the pack documentation on a GitHub hosted runner. |
| pack | Builds the pack on a GitHub hosted runner. |
| test_audio | Builds the Keyword spotting application (audio) for different contexts by using the Arm Compiler for Embedded (AC6), execute each of them by using FVP simulation models, capture the FVP UART output and upload them as artifact. |
| test_video | Builds the Object detection application (video) for different contexts by using the Arm Compiler for Embedded (AC6). |
| test_generic | Builds the a Generic inference application for different contexts by using the Arm Compiler for Embedded (AC6). |
- CMSIS-MLEK Pack available on keil.arm.com/packs.
- tensorflow-lite-micro Pack available on keil.arm.com/packs.
- CMSIS-MLEK Examples a repository with MLEK applications that are configured for evaluation boards and use cases.
- ML Developers Guide for Cortex-M Processors and Ethos-U NPU.
- Arm Virtual Hardware - FVP repository with documentation.
The documentation is generated using MKDocs. Use mkdocs serve to generate the documentation on a local computer.
Contributions are accepted under
Please feel free to raise an issue on GitHub to report misbehavior (i.e. bugs) or start discussions about enhancements. This is your best way to interact directly with the maintenance team and the community. We encourage you to append implementation suggestions as this helps to decrease the workload of the very limited maintenance team.
We will be monitoring and responding to issues as best we can. Please attempt to avoid filing duplicates of open or closed items when possible. In the spirit of openness we will be tagging issues with the following:
- bug – We consider this issue to be a bug that will be investigated.
- wontfix - We appreciate this issue but decided not to change the current behavior.
- enhancement – Denotes something that will be implemented soon.
- future - Denotes something not yet schedule for implementation.
- out-of-scope - We consider this issue loosely related to CMSIS. It might by implemented outside of CMSIS. Let us know about your work.
- question – We have further questions to this issue. Please review and provide feedback.
- documentation - This issue is a documentation flaw that will be improved in future.
- review - This issue is under review. Please be patient.
- DONE - We consider this issue as resolved - please review and close it. In case of no further activity this issues will be closed after a week.
- duplicate - This issue is already addressed elsewhere, see comment with provided references.
- Important Information - We provide essential information regarding planned or resolved major enhancements.