ios

TensorFlow iOS Examples

This folder contains examples of how to build applications for iOS devices using TensorFlow.

Running the Samples using CocoaPod

• You'll need Xcode 7.3 or later.

• There are currently three examples: simple, benchmark, and camera. For now, you can download the sample code by cloning the main tensorflow repository (we are planning to make the samples available as a separate repository later).

• From the root of the tensorflow folder, download Inception v1, and extract the label and graph files into the data folders inside both the simple and camera examples:

mkdir -p ~/graphs
curl -o ~/graphs/inception5h.zip \
 https://storage.googleapis.com/download.tensorflow.org/models/inception5h.zip \
 && unzip ~/graphs/inception5h.zip -d ~/graphs/inception5h
cp ~/graphs/inception5h/* tensorflow/examples/ios/benchmark/data/
cp ~/graphs/inception5h/* tensorflow/examples/ios/camera/data/
cp ~/graphs/inception5h/* tensorflow/examples/ios/simple/data/

• Change directory to one of the samples, download the TensorFlow-experimental pod, and open the Xcode workspace. Observe: installing the pod can take a long time since it is big (~450MB). For example, if you want to run the simple example, then:

cd tensorflow/examples/ios/simple
pod install
open tf_simple_example.xcworkspace # obs, not the .xcodeproj directory

• Run the simple app in the simulator. You should see a single-screen app with a "Run Model" button. Tap that, and you should see some debug output appear below indicating that the example Grace Hopper image in directory data has been analyzed, with a military uniform recognized.

• Run the other samples using the same process. The camera example requires a real device connected. Once you build and run that, you should get a live camera view that you can point at objects to get real-time recognition results.

Troubleshooting

• Make sure you use the TensorFlow-experimental pod (and not TensorFlow).

• The TensorFlow-experimental pod is current about ~450MB. The reason it is so big is because we are bundling multiple platforms, and the pod includes all TensorFlow functionality (e.g. operations). The final app size after build is substantially smaller though (~25MB). Working with the complete pod is convenient during development, but see below section on how you can build your own custom TensorFlow library to reduce the size.

Creating Your own App

• Create your own app using Xcode then add a file named Podfile at the project root directory with the following content:

target 'YourProjectName'
       pod 'TensorFlow-experimental'

• Then you run pod install to download and install the TensorFlow-experimental pod, and finally perform open YourProjectName.xcworkspace and add your code.

• In your apps "Build Settings", make sure to add $(inherited) to sections "Other Linker Flags", and "Header Search Paths".

• That's it. If you want to create your custom TensorFlow iOS library, for example to reduce binary footprint, see below section.

Building the TensorFlow iOS libraries from source

• You'll need Xcode 7.3 or later, with the command-line tools installed.

• Follow the instructions at tensorflow/contrib/makefile under "iOS" to compile a static library containing the core TensorFlow code.

• You should see a single-screen app with a "Run Model" button. Tap that, and you should see some debug output appear below indicating that the example Grace Hopper image has been analyzed, with a military uniform recognized.

• Once you have success there, make sure you have a real device connected and open up the Xcode project in the camera subfolder. Once you build and run that, you should get a live camera view that you can point at objects to get real-time recognition results.

Troubleshooting

If you're hitting problems, here's a checklist of common things to investigate:

• Make sure that you've run the build_all_ios.sh script. This will run download_dependencies.sh,compile_ios_protobuf.sh and compile_ios_tensorflow.sh. (check each one if they have run successful.)

• Check that you have version 7.3 of Xcode.

• If there's a complaint about no Sessions registered, that means that the C++ global constructors that TensorFlow relies on for registration haven't been linked in properly. You'll have to make sure your project uses force_load, as described below.

Creating your Own App from your source libraries

You'll need to update various settings in your app to link against TensorFlow. You can view them in the example projects, but here's a full rundown:

• The compile_ios_tensorflow.sh script builds a universal static library in tensorflow/contrib/makefile/gen/lib/libtensorflow-core.a. You'll need to add this to your linking build stage, and in Search Paths add tensorflow/contrib/makefile/gen/lib to the Library Search Paths setting.

• You'll also need to add libprotobuf.a and libprotobuf-lite.a from tensorflow/contrib/makefile/gen/protobuf_ios/lib to your Build Stages and Library Search Paths.

• The Header Search paths needs to contain:

  • the root folder of tensorflow,
  • tensorflow/contrib/makefile/downloads/protobuf/src
  • tensorflow/contrib/makefile/downloads,
  • tensorflow/contrib/makefile/downloads/eigen, and
  • tensorflow/contrib/makefile/gen/proto.

• In the Linking section, you need to add -force_load followed by the path to the TensorFlow static library in the Other Linker Flags section. This ensures that the global C++ objects that are used to register important classes inside the library are not stripped out. To the linker, they can appear unused because no other code references the variables, but in fact their constructors have the important side effect of registering the class.

• You'll need to include the Accelerate framework in the "Link Binary with Libraries" build phase of your project.

• C++11 support (or later) should be enabled by setting C++ Language Dialect to GNU++11 (or GNU++14), and C++ Standard Library to libc++.

• The library doesn't currently support bitcode, so you'll need to disable that in your project settings.

• Remove any use of the -all_load flag in your project. The protocol buffers libraries (full and lite versions) contain duplicate symbols, and the -all_load flag will cause these duplicates to become link errors. If you were using -all_load to avoid issues with Objective-C categories in static libraries, you may be able to replace it with the -ObjC flag.

Reducing the binary size

TensorFlow is a comparatively large library for a mobile device, so it will increase the size of your app. Currently on iOS we see around a 11 MB binary footprint per CPU architecture, though we're actively working on reducing that. It can be tricky to set up the right configuration in your own app to keep the size minimized, so if you do run into this issue we recommend you start by looking at the simple example to examine its size. Here's how you do that:

• Open the Xcode project in tensorflow/examples/ios/simple.

• Make sure you've followed the steps above to get the data files.

• Choose "Generic iOS Device" as the build configuration.

• Select Product->Build.

• Once the build's complete, open the Report Navigator and select the logs.

• Near the bottom, you'll see a line saying "Touch tf_simple_example.app".

• Expand that line using the icon on the right, and copy the first argument to the Touch command.

• Go to the terminal, type ls -lah and then paste the path you copied.

• For example it might look like ls -lah /Users/petewarden/Library/Developer/Xcode/DerivedData/tf_simple_example-etdbksqytcnzeyfgdwiihzkqpxwr/Build/Products/Debug-iphoneos/tf_simple_example.app

• Running this command will show the size of the executable as the tf_simple_example line.

Right now you'll see a size of around 25 MB, since it's including two architectures (armv7 and arm64). As a first step, you should make sure the size increase you see in your own app is similar, and if it's larger, look at the "Other Linker Flags" used in the Simple Xcode project settings to strip the executable.

After that, you can manually look at modifying the list of kernels included in tensorflow/contrib/makefile/tf_op_files.txt to reduce the number of implementations to the ones you're actually using in your own model. We're hoping to automate this step in the future, but for now manually removing them is the best approach.