Tensorflow iOS Object Detection

An Object Detection application on iOS using Tensorflow and pre-trained COCO dataset models. Video frames are captured and inference is done locally using one of the 3 provided models: ssd_mobilenet_v1_coco, ssd_modelnet_v2_coco, or ssd_inception_v2_coco. Both Swift and Objective-C projects.

Building

• Make sure you have automake and libtool. Using homebrew:

brew install automake libtool

• Clone the tensorflow source repo on GitHub

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

• cd into the tensorflow repo and at a minimum checkout the v1.5.0 tag. The v1.5.0 release has the Makefile support for the following ANDROID_TYPES environment variable. I have tested up to v1.12.0

git checkout origin/r1.11

if you run into this issue: thread-local storage is not supported for the current target, you may need to edit the two files affected. The following one-liners worked for me:

gsed '/ifeq[^,]*,I386)/!b;n;n;n;n;n;s/thread_local//' < ./tensorflow/contrib/makefile/Makefile > foo; mv foo ./tensorflow/contrib/makefile/Makefile

gsed 's/thread_local int per_thread_max_parallism/__thread int per_thread_max_parallism/' < tensorflow/core/util/work_sharder.cc > foo; mv foo ./tensorflow/core/util/work_sharder.cc

You can get gsed using homebrew: brew install gnu-sed

• We need to build the tensorflow components with ANDROID_TYPES_FULL. In the terminal type:

export ANDROID_TYPES="-D__ANDROID_TYPES_FULL__"

• Build the tensorflow libraries for iOS. Go to the root of your newly downloaded tensorflow repo and run:

tensorflow/contrib/makefile/build_all_ios.sh

Go get a coffee. This can take a while. On my macBook it took almost 2 hours.

• Open either the Swift or Objective-C project in this repo and edit the tensorflow.xconfig file to point to the folder where you cloned the tensorflow repo

TENSORFLOW_ROOT=/Users/username/Development/tensorflow

• Compile the xcode project and run. Since we need a camera this will only run on a device.