google-ai-edge/mediapipe
CROSS-PLATFORM ON-DEVICE ML FRAMEWORK
MediaPipe is Google's framework for running ML on-device. Supports mobile platforms, web, desktop, and IoT without sending data to the cloud.
// KEY FEATURES
- Ready solutions for audio, vision, text, and GenAI
- MediaPipe Tasks — cross-platform APIs for deployment
- MediaPipe Studio — visualization and benchmarks in browser
- Local data processing, metrics sent to Google
// DEEP DIVE: HOW IT WORKS
MediaPipe consists of two levels: high-level Solutions (Tasks, Models, Studio) for quick integration and a low-level C++ Framework for building custom pipelines through calculator graphs and data packets. Solutions cover audio, vision, text, decision-making, generative AI, and retrieval, providing pre-trained models and unified APIs for Android, iOS, Python, and Web.
The key difference is strict on-device processing: input data (images, video, text) never leaves the device, which is critical for privacy. At the same time, the framework collects anonymous performance and API usage metrics for sending to Google. The graph-based architecture allows combining calculators into efficient pipelines with minimal overhead.
Requires environment setup for the target platform (Android, iOS, Python, Web). For customization beyond Solutions, you need to work with the C++ Framework and understand packets, graphs, and calculators concepts. Documentation and demos have moved to developers.google.com/edge/mediapipe.
// USE CASES
Best for: For mobile and web developers for on-device ML
- Add gesture recognition to a mobile app without cloud
- Run semantic search on-device via EmbeddingGemma
- Build a custom ML pipeline for real-time pose tracking
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram