Skip to content
Branch: master
Find file History

README.md

TensorFlow Lite sample applications

The following samples demonstrate the use of TensorFlow Lite in mobile applications. Each sample is written for both Android and iOS.

Image classification

This app performs image classification on a live camera feed and displays the inference output in realtime on the screen.

Samples

Android image classification

iOS image classification

Raspberry Pi image classification

Object detection

This app performs object detection on a live camera feed and displays the results in realtime on the screen. The app displays the confidence scores, classes and detected bounding boxes for multiple objects. A detected object is only displayed if the confidence score is greater than a defined threshold.

Samples

Android object detection

iOS object detection

Raspberry Pi object detection

Speech command recognition

This application recognizes a set of voice commands using the device's microphone input. When a command is spoken, the corresponding class in the app is highlighted.

Samples

Android speech commands

iOS speech commands

Gesture classification

This app uses a model to classify and recognize different gestures. A model is trained on webcam data captured using a web interface. The model is then converted to a TensorFlow Lite model and used to classify gestures in a mobile application.

Gesture components

Web app

First, we use TensorFlow.js embedded in a web interface to collect the data required to train the model. We then use TensorFlow.js to train the model.

Web gesture classification

Conversion script

The model downloaded from the web interface is converted to a TensorFlow Lite model.

Conversion script (available as a Colab notebook).

Mobile apps

Once we have the TensorFlow Lite model, the implementation is very similar to the Image classification sample.

Samples

Android gesture classification

iOS gesture classification

Model personalization

This app performs model personalization on a live camera feed and displays the results in realtime on the screen. The app displays the confidence scores, classes and detected bounding boxes for multiple objects that were trained in realtime.

Samples

Android Model Personalization

You can’t perform that action at this time.