Preface

Recently, terms like machine learning, artificial intelligence, and deep learning have become extremely popular. In this post, I use deep learning to build a simple Android app that can recognize different types of flowers. Part 1 trains an image model, which you can already use for classification on your computer. Part 2 puts the trained model into an Android project so you can do recognition on a phone.

Part 1

Environment Setup

Anaconda

Anaconda lets you create virtual environments to run Python. The benefit is that you can create many different environments for testing, and when using Anaconda, pip does not require root privileges.

$ wget https://repo.continuum.io/archive/Anaconda3-4.4.0-Linux-x86_64.sh
$ bash Anaconda3-4.4.0-Linux-x86_64.sh

Python & TensorFlow

Create a virtual environment. Since I used Python 2.7 with TensorFlow 1.1, I named the environment py2t1.1.

$ conda create -n py2t1.1 python=2.7

After installing, enter the virtual environment. To activate, use activate:

 $ source activate py2t1.1
 (py2t1.1)$  # it will look like this

To deactivate, use deactivate:

 $ source deactivate

Next, install tensorflow. Here I use TensorFlow 1.1 (Python 2.7, CPU). If you need other versions, you can find them here.

 (py2t1.1)$ pip install --ignore-installed --upgrade \
 https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.1.0-cp27-none-linux_x86_64.whl

Now the environment is ready.

Training the Model

Preparation

At this point, you can be inside the virtual environment or not. You only need to enter the environment when you run TensorFlow with Python.

Create a folder:

$ mkdir flower_classification
$ cd flower_classification

Download the flower photos:

curl -O http://download.tensorflow.org/example_images/flower_photos.tgz
tar xzf flower_photos.tgz

Download retrain.py. It uses an Inception v3 image recognition model and adds a new final layer, then retrains the model to classify flowers.

$ curl -O https://raw.githubusercontent.com/tensorflow/tensorflow/r1.1/tensorflow/examples/image_retraining/retrain.py

Start Training

First, activate the environment:

$ source activate py2t1.1

$ python retrain.py \
  --bottleneck_dir=bottlenecks \
  --model_dir=inception \
  --learning_rate=0.5 \
  --summaries_dir=training_summaries/long \
  --output_graph=retrained_graph.pb \
  --output_labels=retrained_labels.txt \
  --image_dir=flower_photos

After training, it generates retrained_graph.pb and retrained_labels.txt. You can use these two files to classify new photos.

Classifying Photos

Download label_image.py, which is the script for classification:

$ curl -L https://goo.gl/3lTKZs > label_image.py

Run classification:

$ python label_image.py 要辨識的花朵照片.jpeg

The output will look like this:

daisy (score = 0.99071)
sunflowers (score = 0.00595)
dandelion (score = 0.00252)
roses (score = 0.00049)
tulips (score = 0.00032)

Part 2

Preparation

tensorflow-for-poets-2 is a sample Android app project from Google Codelab.

$ cd ~/
$ git clone https://github.com/googlecodelabs/tensorflow-for-poets-2
$ cd tensorflow-for-poets-2

Copy the entire flower classification folder flower_classification (trained earlier) into the project:

$ cp -r ~/flower_classification .

Processing the Model

Optimize:

$ python -m tensorflow.python.tools.optimize_for_inference \
  --input=flower_classification/retrained_graph.pb \
  --output=flower_classification/optimized_graph.pb \
  --input_names="Cast" \
  --output_names="final_result"

Quantize:

$ python -m scripts.quantize_graph \
  --input=flower_classification/optimized_graph.pb \
  --output=flower_classification/rounded_graph.pb \
  --output_node_names=final_result \
  --mode=weights_rounded

Put rounded_graph.pb and retrained_labels.txt into the android/assets folder:

$ cp flower_classification/rounded_graph.pb flower_classification/retrained_labels.txt android/assets/

Installing Android Studio

https://developer.android.com/studio/index.html

Opening the Android Project

Open Android Studio, choose “Open an existing Android Studio project”, and select the folder tensorflow-for-poets-2/android. Then it will ask whether to use Gradle—choose “ok”. After Gradle finishes syncing, you can build the project, and finally install the APK onto your phone.

References

https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/index.html#0
https://codelabs.developers.google.com/codelabs/tensorflow-for-poets-2/#0