Tensorflow Lite Micro Software Package

1. Introduction

This software package is the Tensorflow Lite embedded inference framework Tensorflow Lite Micro software package for the RT-Thread ecosystem . Through this software package, the end deployment task of the deep learning model trained based on the Tensorflow Lite framework can be realized in the embedded system.

1.1 Directory structure

Name	Description
docs	Document
examples	Tensorflow Lite Micro offical voice demo
fixedpoint	Fixed-point quantization library required by Tensorflow Lite Micro library
flatbuffers	Model interpretation library flatbuffer required by Tensorflow Lite Micro library
ruy	Matrix acceleration library required by Tensorflow Lite Micro library ruy
tensorflow	Tensorflow Lite Micro library

1.2 License

Tensorflow Lite Micro package complies with the LGPLv2.1 license, please refer to the LICENSE file for details.

1.3 Dependency

RT-Thread 3.0+

2. How to open Tensorflow Lite Micro

To use Tensorflow Lite Micro package, you need to select it in the package manager of RT-Thread. The specific path is as follows:

RT-Thread online packages
    miscellaneous packages --->
        [*] Tensorflow Lite Micro: a lightweight deep learning end-test inference framework for RT-Thread operating system.

Then let the RT-Thread package manager automatically update, or use the pkgs --update command to update the package to the BSP.

3. Use Tensorflow Lite Micro

After successfully downloading the Tensorflow Lite Micro package:

Change the package folder name of TensorflowLiteMicro_xxx (where xxx is the package version number) downloaded under the packages file to TensorflowLiteMicro (that is, remove the version number, so as not to affect compilation)

Function configuration through menuconfig, the configuration options in menuconfig are:

RT-Thread online packages
miscellaneous packages --->
    [*] Tensorflow Lite Micro: a lightweight deep learning end-test inference framework for RT-Thread operating system.
        Version(latest) --->
        Select Offical Example(Enable Tensorflow Lite Micro aduio example) --->

Among them, there are two options in Select Offical Example:

(X) Enable Tensorflow Lite Micro audio example
( ) No Tensorflow Lite Micro example

Among them, audio example is to implement the official audio demo, No example does not integrate the example file, only uses the Tensorflow Lite Micro standard framework. For the precautions of the menucofing option, please refer to the 4. Precautions section

The whole framework of Tensorflow Lite Micro has more complicated functions and many APIs. Please refer to introduction.md first, and then use user-guide.md to learn the basic deployment process, on this basis, consider the issue of custom development.
The API manual can visit this link, which provides the current support for operators
More documents are located under /docs, Be sure to check before use

4. Matters needing attention

If the audio example option is selected in the menuconfig, the software package will use the self main function, and the user needs to manually delete all main functions except packages/TensorflowLiteMicro/example/audio_main.cc
If No example is selected, the system does not have main function, users can design the main function according to their own needs to call the Tensorflow Lite Micro framework
This software package will occupy approximately 480KB Flash space and 92KB RAM space after compilation, so it is not recommended for MCU systems with few hardware resources to use this software package
This software package occupies 16KB RAM space when running, and the built-in voice recognition case occupies a total of 22KB memory when running. **You need to modify the size of the main function stack and the memory management algorithm through menuconfig! **
This software package currently only runs successfully on the Raspberry Pi 4 platform, other platforms have yet to be tested. Raspberry Pi 4 porting repositories link: https://github.com/QingChuanWS/raspi4-tfliteMicro

5. Contact & Thanks

Maintenance: QingChuanWS
Homepage: https://github.com/QingChuanWS

README_en.md 4.5 KB لینک دائمی تاريخچه خام