A full build of TensorFlow for iOS. Unofficial. Latest is v2.0.9 based on Tensorflow r2.0, v2.0.3 and built with Xcode 12.3 (12C33).
Framework targets simluator and arm64 devices only (iOS 12.0+) with full support for training MobileNetV2 models on device.
This repository contains only the required header files for a portable build of tensorflow for iOS. The tensorflow static library must be added to the root directory of the framework.
To create this framework, first run build_all_ios.sh from tensorflow/contrib/makefile on branch r2.0.doc.ai in our custom tensorflow repository and then run create_full_ios_frameworks.sh in the same directory.
The following static library must be added to the root directory of this framework. Rename it simply to tensorflow
You may also run the download_libs.sh script in this repository to download and add the tensorflow static library to the framework.
The following additional static libraries must be added as dependencies to any project which uses this framework:
The framework must be added to the "Embedded Binaries" and "Linked Frameworks and Libraries" section of your Xcode project, in the General build setting.
Add the protobuf and nsync static libraries to the "Linked Frameworks and Libraries".
Add -force_load $(PROJECT_DIR)/tensorflow.framework/tensorflow to the Other Linker Flags in the projects Build Settings.
This framework is wrapped in the TensorIOTensorFlow pod. The cocoapod is the simplest way to include this framework in your project. It vends the tensorflow, protobuf, and nsync libraries as well as all the required headers.
Include the pod with:
pod 'TensorIOTensorFlow'
Learn more about compiling tensorflow on your machine.
Learn more about customizing the ops in your build.
doc.ai
TensorIOTensorFlow is available under the Apache 2.0 license. See the LICENSE file for more info.