This is a camera app that continuously detects the objects (bounding boxes, classes, and confidence) in the frames seen by your device's back camera, in an image imported from the device gallery, or in a video imported by the device gallery, with the option to use a quantized MobileNetV2 EfficientDet Lite 0, or EfficientDet Lite2 model.

The model files are downloaded by a Gradle script when you build and run the app. You don't need to do any steps to download TFLite models into the project explicitly unless you wish to use your own models. If you do use your own models, place them into the app's assets directory.

This application should be run on a physical Android device to take advantage of the physical camera, though the gallery tab will enable you to use an emulator for opening locally stored files.

• The Android Studio IDE. This sample has been tested on Android Studio Dolphin.

• A physical Android device with a minimum OS version of SDK 24 (Android 7.0 - Nougat) with developer mode enabled. The process of enabling developer mode may vary by device. You may also use an Android emulator with more limited functionality.

• 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 mediapipe/examples/object_detection/android directory. Click OK. You may be asked if you trust the project. Select Trust.

• 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.

Downloading, extraction, and placing the models into the assets folder is managed automatically by the download.gradle file.