This post documents my experience reproducing the results in blog https://heartbeat.fritz.ai/neural-networks-on-mobile-devices-with-tensorflow-lite-a-tutorial-85b41f53230c on an Android Phone. The tutorial is about developing an Andoid app for image classification using TensorFlow Lite.

My machine is Lenovo Ideapad Flex 5-1570 running Windows 10 Home.

  1. Install Anaconda (Python 3.7 version) from Anaconda website (630MB).

  2. Click Start Menu -> Anaconda Prompt to open terminal.

  3. In terminal, install TensorFlow by 
    pip install tensorflow
  4. In terminal, update NumPy to v1.6: 
    pip install numpy —upgrade

    The numpy coming with TensorFlow is v1.15, which will cause error when training.

  5. Install Git for Windows (if not yet).

  6. Git clone the code files. 
    git clone https://github.com/googlecodelabs/tensorflow-for-poets-2

  7. Download training images and unzip to the location designated in tutorial.

  8. Train mobinenet as instructed in tutorial. 
    python scripts\retrain.py --output_graph=tf_files\retrained_graph.pb --output_labels=tf_files\retrained_labels.txt--image_dir=tf_files\flower_photos
--architecture=mobilenet_1.0_224 --summaries_dir tf_files\training_summaries\mobilenet_1.0_244
  9. Convert to TF Lite format 
    toco --graph_def_file=tf_files\retrained_graph.pb --output_file=tf_files\optimized_graph.lite  --output_format=TFLITE --input_shape=1,224,224,3 --input_array=input --output_array=final_result
--inference_type=FLOAT --inference_input_type=FLOAT --input_format=TENSORFLOW_GRAPHDEF

  10. Download and install Android Studio (taking a while).

  11. Follow instructions in tutorial to update the graph.lite and labels.txt files in folder android\tflite\app\src\main\assets.

  12. Follow instructions in tutorial to open the project in \tflite folder. Fist time building the app, there will be a few errors when building the project. Just follow the error messages to delete a few lines. Build the APK file as instructed.

  13. Connect a device (BLUE ADVANCE 5.5 HD in my case). Turn on the Developer Options by following instructions at https://developer.android.com/studio/debug/dev-options. Then I was able to run the demo app on the device.