MediaPipe Samples

This Android application demonstrates how to compare the similarity between two texts using MediaPipe Tasks. It supports two model options:

1. Mobile Bert: Uses mobilebert_embedding_with_metadata.tflite.
2. Average word-embedding: Uses regex_embedder_with_metadata.tflite.

Model Management:

• Automatic: A Gradle script (download.gradle) automatically downloads and extracts the models into the app's assets directory during the build process.
• Custom Models: If you want to use your own models, manually place the .tflite files into the app's assets directory.

MediaPipe Solutions provides tools for on-device machine learning development and deployment across multiple platforms. It is composed of three main components:

• MediaPipe Tasks (low-code): Used to create and deploy custom end-to-end ML solution pipelines.
• MediaPipe Model Maker (low-code): Used to create custom ML models from advanced solutions.
• MediaPipe Studio (no-code): A platform to create, evaluate, debug, benchmark, prototype, and deploy advanced production-level solutions.

This Android demo is a camera application that performs continuous object segmentation on frames captured by the device's back camera. It also supports segmenting images and videos imported from the device gallery.

Key Features:

• Real-time Camera Segmentation: Uses the device's back camera for live segmentation.
• Gallery Support: Allows users to select images or videos from the device gallery for segmentation.
• Automated Model Management: Model files are automatically downloaded and placed in the assets directory via a Gradle script during the build process.