Cross-Platform TFLite Object Detection

This is a sample application of running object detection using TensorFlow Lite in c++

It is part of this blog post series.

This repo contains sample application in Android and iOS.

Setup

TensorFlow Lite libraries

This project is dependent on TensorFlow Lite lib and headers, you can either build them yourself (as explained) in the blog post above, or by running download-tf-dist.sh which downloads prebuilt libraries from this repo of mine.

In any case, the project expect to have a folder structure like so:

Headers:

tflite-dist/include/flatbuffers - Flatbuffers headers tflite-dist/include/tensorflow/lite - tflite headers

TensorFlow Lite shared object library:

tflite-dist/libs/andoird/arm64-v8a/libtensorflowlite.so tflite-dist/libs/andoird/armeabi-v7a/libtensorflowlite.so tflite-dist/libs/andoird/x86/libtensorflowlite.so tflite-dist/libs/andoird/x86_64/libtensorflowlite.so tflite-dist/libs/ios/libtensorflow-lite.a

OpenCV

This project makes use of OpenCV (tested against version 4.0.1)

OpenCV for Android

• Download OpenCV Android SDK from here
• Extract the zip file.
• Define global environment variable OPENCV_ANDROID pointing to the root folder of the opencv android sdk (by "global environment variable" the meaning is that it will be available for Android Studio).
• This environment variable is used by app/CMakeLists.txt as follow: set(OpenCV_DIR $ENV{OPENCV_ANDROID}/sdk/native/jni) so double check it points to the correct location

For a complete guide see this blog post

OpenCV for iOS

Managed as a pod, just pod install

Object Detection Model

Note that this repo contains a tflite object detection model (SSD MobileNET V1 published by Google, link to original) so make sure the folder tflite-models is available as it is being referenced as an embedded resource by both apps.

Sample output

License

MIT