TensorFlow Lite Gesture Classification
Overview
This is a camera app that continuously classifies the gesture in the frames seen by your device's front camera (changing it back for STB). These instructions walk you through building and running the app on an Android device.
The model files are downloaded via Gradle scripts when you build and run the app. You don't need to do any steps to download TFLite models into the project explicitly.
We trained the model with the following hand gesture images.
This application should be run on a physical Android device.
Build the app using Android Studio
Prerequisites
-
The Android Studio IDE (Android Studio 2021.2.1 or newer). This sample has been tested on Android Studio Chipmunk.
-
A physical Android device with a minimum OS version of SDK 23 (Android 6.0 - Marshmallow) with developer mode enabled. The process of enabling developer mode may vary by device.
Building
-
Open Android Studio. From the Welcome screen, select Open an existing Android Studio project.
-
From the Open File or Project window that appears, navigate to and select the tensorflow-lite/examples/gesture_classification/android directory. Click OK.
-
If it asks you to do a Gradle Sync, click OK.
-
With your Android device connected to your computer and developer mode enabled, click on the green Run arrow in Android Studio.
Customize your model
-
Read the following doc to generate TFLite model file.
-
Add the metadata to the tflite model using this Colab notebook
-
Copy the downloaded model into app/src/main/assets folder. Make sure the model is named model_metadata.tflite for this sample.