Abstract:

This video provides a comprehensive guide on installing TensorFlow Lite (TFLite) C++ on Windows. It outlines the step-by-step process, starting with the installation of essential prerequisites such as Python, Bazel, MSYS2, and Visual C++ build tools. The tutorial then demonstrates how to download the TensorFlow source, configure the build environment, compile the TFLite library using Bazel, and finally, verify the successful installation by building and running a minimal C++ example with a TFLite model.

Installing TensorFlow Lite C++ on Windows: A Step-by-Step Guide

• 0:00 Introduction: The video introduces a tutorial on installing TensorFlow Lite (TFLite) C++ on the Windows operating system.
• 0:08 Python Installation: The first step is to download and install the latest Python version from the official Python website.
• 0:42 Install Python Packages: Essential Python packages like numpy, keras_preprocessing, absl_py, wheel, setuptools, and pip are installed using pip.
• 0:56 Bazel Installation: Bazel, a build tool similar to CMake, is installed. Specifically, version 3.7.2 is downloaded as a Windows executable, renamed to "bazel," and its path is added to the system's environmental variables.
• 0:1:43 MSYS2 Tool Setup: The MSYS2 tool is installed, and its usr/bin folder is added to the system's environmental variables to provide a Unix-like environment on Windows.
• 0:2:30 Install Git, Patch, Unzip: Within the MSYS2 environment, git, patch, and unzip libraries are installed using the pacman package manager.
• 0:2:50 Visual C++ Build Tools: Visual C++ build tools are required, specifically the "Desktop C++" development and MSVC compilers, which can be obtained through Visual Studio Community Edition.
• 0:3:06 Download TensorFlow Source: The latest zipped version of the TensorFlow source code is downloaded from its release page.
• 0:3:45 Configure Build: Inside the extracted TensorFlow directory, the ./configure script is run to set up the build. Key configurations include specifying the Python path, opting out of ROCm and GPU support (as TFLite doesn't natively support NVIDIA GPUs on Windows for C++), enabling optimization settings, reducing compilation time, and skipping Android builds.
• 0:4:20 Build TensorFlow Lite: The TFLite library is compiled using the Bazel command: bazel build -c opt //tensorflow/lite:tensorflowlite.
• 0:4:43 Locate Built Files: Upon successful compilation, the tensorflowlite.lib and tensorflowlite.dll files can be found in the bazel-bin/tensorflow/lite directory.
• 0:5:02 Verify Installation with Example Project: A GitHub repository (karthik_a_tflite) containing a minimal TFLite C++ example is cloned. This project demonstrates the necessary include directories, libraries (flatbuffers.lib, tensorflowlite.lib), and how to copy tensorflowlite.dll to the executable directory.
• 0:6:31 Build and Run Example: The example project is built using CMake and Visual Studio. The resulting executable (tf_flight_check.exe) is then run from the terminal, taking a TFLite model (e.g., mobilenet_ssd_tflite_model) as an argument, to confirm that the TFLite C++ installation is working correctly.