google-ai-edge/mediapipe

MediaPipe – On‑device ML pipelines made easy

What it is – MediaPipe is an open‑source framework from Google for building fast, on‑device machine‑learning pipelines (called graphs). It ships ready‑made Solutions such as hand‑tracking, pose estimation, object detection, text‑ and audio‑classifiers, and lets developers plug these into Android, iOS, web, desktop or other edge devices.

Key pieces

• MediaPipe Framework – low‑level C++ library (with Android/iOS bindings) that lets you assemble calculators that process packets of data in a directed graph. This is the engine for custom perception pipelines.
• MediaPipe Tasks – higher‑level, cross‑platform APIs (Python, Java, JavaScript, C++) that expose the ready‑made solutions as simple functions you can call from your app.
• Model Maker – a tool to fine‑tune the bundled pre‑trained models on your own data.
• MediaPipe Studio – a browser‑based visualiser for building, debugging and benchmarking graphs.

Why it matters – All inference runs locally, so no user data leaves the device, which is important for privacy‑sensitive apps. The framework is heavily optimised for mobile/edge hardware, giving real‑time performance for vision, audio and text tasks.

Typical use‑cases

• Adding background‑blur or virtual‑backgrounds in video‑conferencing apps.
• Real‑time hand‑tracking for AR/VR, sign‑language interfaces, prosthetic control.
• Pose‑estimation for fitness, gaming, or motion capture.
• On‑device object detection for robotics, smart cameras, or retail.

Getting started – Follow the quick‑start guides on the official docs for the platform you target (Android, Web, Python). The repo also contains example apps and the source of the legacy solutions.

Community & support – Slack channel, Google Group, Stack Overflow tag, and an "Awesome MediaPipe" curated list.

License – Open source under the Apache 2.0 licence.