[Raspberry Pi] Install TensorFlow Lite on Raspberry Pi 4

Build TensorFlow Lite Static Library on Raspberry Pi 4

I found a very useful resource about installing TensorFlow Lite 2.* on Raspberry Pi 4 as follows: https://qengineering.eu/install-tensorflow-2-lite-on-raspberry-pi-4.html

It's about to build TensorFlow Lite static library in version 2.4 :

# the tools needed

$ sudo apt-get install cmake curl
# download the latest TensorFlow version (2.4.0)
$ wget -O tensorflow.zip https://github.com/tensorflow/tensorflow/archive/v2.4.0.zip
# unpack and give the folder a convenient name
$ unzip tensorflow.zip
$ mv tensorflow-2.4.0 tensorflow
$ cd tensorflow
# get the dependencies
$ ./tensorflow/lite/tools/make/download_dependencies.sh
# run the C++ installation (± 25 min)
$ ./tensorflow/lite/tools/make/build_rpi_lib.sh

pi@raspberrypi:~/Downloads/tensorflow-2.4.0/tensorflow/lite/tools/make/gen$ tree -L 3
.
└── rpi_armv7l
    ├── bin
    │   ├── benchmark_model
    │   ├── benchmark_model_performance_options
    │   └── minimal
    ├── lib
    │   ├── benchmark-lib.a
    │   └── libtensorflow-lite.a
    └── obj
        └── tensorflow

Or, if you want to use cmake to do so, please follow the steps:

Step 1. Install CMake tool

$ sudo apt-get install cmake

Step 2. Clone TensorFlow repository

$ git clone https://github.com/tensorflow/tensorflow.git tensorflow_src

Note: If you're using the TensorFlow Docker image, the repo is already provided in /tensorflow_src/.

Step 3. Create CMake build directory

$ mkdir tflite_build

$ cd tflite_build

After that, run the following commands.

$ cmake -DTFLITE_ENABLE_XNNPACK=OFF ../tensorflow_src/tensorflow/lite

$ cmake --build . -j4

Build TensorFlow Lite Shared Library on Raspberry Pi 4 using Bazel-on-ARM project

Bazel is an open-source build tool from Google, used to build projects such as TensorFlow. Raspberry Pi OS (Raspbian) doesn't have a package for Bazel, and the Bazel project doesn't provide a binary for armhf.

Please install bazel first based on Bazel on ARM on Raspberry Pi

$ git clone https://github.com/koenvervloesem/bazel-on-arm
$ cd bazel-on-arm
$ sudo make requirements
$ make bazel
#or If you want to build a specific Bazel version, run the build script with the version number as an argument:
$ env EXTRA_BAZEL_ARGS="--host_javabase=@local_jdk//:jdk" bash ./scripts/build_bazel.sh 3.7.2
#You can install it to /usr/local/bin with:
sudo make install

There are pre-build bazel binary files in Release: https://github.com/koenvervloesem/bazel-on-arm/releases

Build TensorFlow Lite ( is not successful )
$ bazel build -c opt //tensorflow/lite/c:libtensorflowlite_c.so

Prebuilt binary for TensorflowLite

TensorflowLite-bin (Important)

Prebuilt binary for TensorflowLite's standalone installer. For RaspberryPi. I provide a FlexDelegate, MediaPipe Custom OP, XNNPACK and XNNPACK Multi-Thread PythonAPI enabled binary. 

If you want the best performance with RaspberryPi4/3, install Ubuntu 18.04+ aarch64 (64bit) instead of Raspbian armv7l (32bit). The official Tensorflow Lite is performance tuned for aarch64. On aarch64 OS, performance is about 4 times higher than on armv7l OS

Python3.7 - Buster

$ wget https://github.com/PINTO0309/TensorflowLite-bin/releases/download/v2.8.0/tflite_runtime-2.8.0-cp37-none-linux_armv7l.whl

$ sudo pip3 install tflite_runtime-2.8.0-cp37-none-linux_armv7l.whl

The successful way to build and install TensorFlow Lite on Raspberry Pi 4:

Use Bazelisk to install Bazel 
https://bazel.build/install/bazelisk

#First, we need to install bazelisk, and make sure you get the correct CPU arch and 64/32 bit for node.js 
$ sudo npm install -g @bazel/bazelisk
#We can directly build bazel from source: ( it works )

$ cd tensorflow
$ ./configure
$ bazel build -c opt //tensorflow/lite/c:libtensorflowlite_c.so
$ sudo cp -rf bazel-bin/
$ sudo cp -rf bazel-bin/tensorflow/lite/c/* /usr/local/lib
$ sudo mkdir -p /usr/local/include/tensorflow/lite/c
$ 
$ sudo mkdir -p /usr/local/include/tensorflow/lite/c
$ sudo cp -rf tensorflow/lite/c/* /usr/local/include/tensorflow/lite/c
$ sudo ldconfig

Further Reading:
Building Bazel and TensorFlow 2.x on aarch64 ( ARM Board )
https://community.arm.com/arm-community-blogs/b/ai-and-ml-blog/posts/building-bazel-and-tensorflow-2-x-on-aarch64