Skip to content

Latest commit

 

History

History
63 lines (48 loc) · 2.55 KB

tflm-guide.md

File metadata and controls

63 lines (48 loc) · 2.55 KB

Guide to Integrate TensorFlow Lite Micro (TFLM) into STM32 with CMake

Implementing TensorFlow Lite Micro on STM32 involves several steps, each detailed below.

Step 1: Building a Library from TensorFlow Lite Source Code

To integrate TensorFlow Lite Micro, you need its source code, which is available on the official TensorFlow Lite Micro GitHub repository. Follow these steps:

  1. Clone the repository:
   git clone https://github.com/tensorflow/tflite-micro.git
  1. Go to the tflite-micro directory
cd tflite-micro
  1. Compiling the Library
make -f tensorflow/lite/micro/tools/make/Makefile TARGET=cortex_m_generic TARGET_ARCH=cortex-m4 microlite

if your Microcontroller has FPU then use TARGET_ARCH=cortex-m4+fp also you can change the TARGET_ARCH according your requirements

Step 2: Add The Compiled library to your project with CMake

If you successfully followed those steps you already have a library named libtensorflow-microlite.a in your {tflite_dir}/gen directory now you have to add this into your project with the help of CMake

link_directories(${CMAKE_SOURCE_DIR}/path-to-lib-directory) # Link your Library Directory and Make it accessible

target_link_libraries(${PROJECT_NAME} PRIVATE your-lib-dir-path/libtensorflow-microlite.a) # Link TensorFlow Lite Micro library

Step 3: Generate the TFLM Tree

Now, you have to add all the necessary Headers to your project but for that you have to generate the TFLM tree because tensorflow lite micro comes with many dependencies which it needs to fetch from the internet and then it will provide you the files which you can include in your project

To generate the TFLM Tree use

python3 {your-tflm-dir}/tensorflow/lite/micro/tools/project_generation/create_tflm_tree.py /your/desired/path/tflm-tree

Step 4: Add all the necessary Headers to your project with CMake

target_include_directories(${PROJECT_NAME} PUBLIC
    ${CMAKE_SOURCE_DIR}/tflm/third_party/flatbuffers/include
    ${CMAKE_SOURCE_DIR}/tflm/third_party/gemmlowp
    ${CMAKE_SOURCE_DIR}/tflm
    ${CMAKE_SOURCE_DIR}/tflm/tensorflow/lite
)

Note: these files are included according to my project setup you have to modify according to your project structure

Few Additional Tips

  1. Change your filename from main.c to main.cc or main.cpp
  2. You may have to use third_party tools like flatbuffers or kissfft then follow step 4.
  3. You can delete all the source files since we already have compiled the library according to our MCU Architecture