jackweiwang / android_tflite

GPU Accelerated TensorFlow Lite applications on Android NDK. Higher accuracy face detection, Age and gender estimation, Human pose estimation, Artistic style transfer

GPU Accelerated TensorFlow Lite applications on Android NDK.

Run and measure the performance of TensorFlow Lite GPU Delegate on Android NDK.

1. Applications

Blazeface

• Lightweight Face Detection.
  

DBFace

• Higher accurate Face Detection.
  

Age Gender Estimation

• Detect faces and estimage their Age and Gender
• based on pretrained model of https://github.com/yu4u/age-gender-estimation
  

Image Classification

• Image Classfication using Moilenet.
  

Object Detection

• Object Detection using MobileNet SSD.
  

Hair Segmentation

• Hair segmentation and recoloring.
  

3D Handpose

• 3D Handpose Estimation from single RGB images.
  

Iris Detection

• Eye position estimation by detecting the iris.
  

Posenet

• Pose Estimation.
  

Depth Estimation (DenseDepth)

• Depth Estimation from single images.
  
• based on pretrained model of https://github.com/ialhashim/DenseDepth

Semantic Segmentation

• Assign semantic labels to every pixel in the input image.
  

Selfie to Anime

• Generate anime-style face image.
  
• based on pretrained model of https://github.com/margaretmz/Selfie2Anime-with-TFLite
  

Anime GAN

• Transform photos into anime style images.
  
• based on pretrained model of https://github.com/TachibanaYoshino/AnimeGANv2
  

U^2-Net portrait drawing

• Human portrait drawing by U^2-Net.
  

Artistic Style Transfer

• Create new artworks in artistic style.
  

MIRNet

• Enhance low-light images upto a great extent.
  
• based on pretrained model of https://github.com/sayakpaul/MIRNet-TFLite

2. How to Build & Run

2.1 setup environment

• Download and install Android NDK.

$ mkdir ~/Android/
$ mv ~/Download/android-ndk-r20b-linux-x86_64.zip ~/Android
$ cd ~/Android
$ unzip android-ndk-r20b-linux-x86_64.zip

• Download and install bazel.

$ wget https://github.com/bazelbuild/bazel/releases/download/3.1.0/bazel-3.1.0-installer-linux-x86_64.sh
$ chmod 755 bazel-3.1.0-installer-linux-x86_64.sh
$ sudo ./bazel-3.1.0-installer-linux-x86_64.sh

2.2 build TensorFlow Lite library and GPU Delegate library

• run the build script to build TensorFlow Library

$ mkdir ~/work
$ git clone https://github.com/terryky/android_tflite.git
$ cd android_tflite/third_party/
$ ./build_libtflite_r2.4_android.sh

(Tensorflow configure will start after a while. Please enter according to your environment)


$ ls -l tensorflow/bazel-bin/tensorflow/lite/
-r-xr-xr-x  1 terryky terryky 3118552 Dec 26 19:58 libtensorflowlite.so*

$ ls -l tensorflow/bazel-bin/tensorflow/lite/delegates/gpu/
-r-xr-xr-x 1 terryky terryky 80389344 Dec 26 19:59 libtensorflowlite_gpu_delegate.so*

2.3 Download the needed assets

$ cd ~/work/android_tflite
$ ./download_all_assets.sh

2.4 Build Android Applications

• Download and install Android Studio.
• Start Android Studio.

$ cd ${ANDROID_STUDIO_INSTALL_DIR}/android-studio/bin/
$ ./studio.sh

• Install NDK 20.0 by SDK Manager of Android Studio.
• Open application folder (eg. ~/work/android_tflite/tflite_posenet).
• Build and Run.

3. Tested Environment

Host PC	Target Device
x86_64	arm64-v8a
Ubuntu 18.04.4 LTS	Android 9 (API Level 28)
Android NDK r20b

4. Related Articles

• Android NDK 環境で TensorFlow Lite GPU Delegate を使う方法 (Qiita)

5. Acknowledgements

• https://github.com/google/mediapipe
• https://github.com/yu4u/age-gender-estimation
• https://github.com/TachibanaYoshino/AnimeGANv2
• https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/python_demos/human_pose_estimation_3d_demo
• https://github.com/ialhashim/DenseDepth
• https://github.com/MaybeShewill-CV/bisenetv2-tensorflow
• https://github.com/margaretmz/Selfie2Anime-with-TFLite
• https://github.com/NathanUA/U-2-Net
• https://tfhub.dev/sayakpaul/lite-model/east-text-detector/int8/1
• https://github.com/PINTO0309/PINTO_model_zoo