This is an android app for object detection using TensorFlow Lite on a mobile device. I have trained my own model using the famous mnist dataset, and saved in .h5 format. Then I converted it to a tflite model for use in the app using this command:
tflite_convert --output_file=mnist_model.tflite --keras_model_file=mnist_model.h5
The main problem with deployinmg machine learning models to mobile devices is the limited memory computation power. TensorFlow Lite is a lightweight package that solves these issues. Here is a screen recording of the app.
TensorFlow Lite is not designed to train a model, the model can be trained on a higher power device. Then, the pretrained model can be converted to a TensorFlow Lite format (.tflite), which has a smaller footprint than can be easily run on a mobile or other embedded devices for classification, regresion or other such tasks. The model (.tflite) file and the class labels (.txt) file need to be placed in the assets folder of the android app.
In the app, there is a canvas on top where the user draws the digit. The app then invokes the digit recognition model. The output is given in the form of a toast. The digit recogniser is encapsulated by DigitsClassifier which uses the TensorFlow Lite Interpreter.
import org.tensorflow.lite.Interpreter
It is very easy to initialize the interpreter with the model:
private Interpreter tflite; tfLite = new Interpreter(loadModelFile(assetManager));
The model is called by the below line, with a grayscale input of size 28x28:
tfLite.run(byteBuffer, outputClasses);
Digit 2
Digit 7
Digit 8
