mediapipe

What it is and what it does

MediaPipe is Google's framework for deploying machine learning models on-device across mobile, web, desktop, and edge platforms. It bundles pre-trained models for common tasks—object detection, pose estimation, hand tracking, face recognition, text classification, and audio analysis—alongside cross-platform APIs to integrate them into applications. All input data (images, video, text, audio) stays on the device; only performance metrics are sent to Google.

The package targets developers and researchers who need to add ML capabilities without building models from scratch or managing cloud inference. It depends on numpy, opencv-contrib-python, matplotlib, sounddevice, flatbuffers, absl-py, and certifi. Installation requires a compatible wheel for your platform (arm64 or x86_64 Linux, macOS 11+, or Windows), and you must supply or download pre-trained model files separately. The framework is actively maintained and in alpha status, with documentation and setup guides available for Python, Android, web, and iOS.

Use it for:

• Build real-time pose or hand tracking in mobile apps or desktop applications without cloud inference.
• Add object detection or face recognition to video processing pipelines on edge devices.
• Classify text or audio input on-device for privacy-sensitive applications.
• Prototype ML features quickly using pre-trained models before customizing with MediaPipe Model Maker.
• Deploy ML inference in IoT or embedded systems where latency and privacy are critical.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

MediaPipe provides pre-built machine learning models and cross-platform APIs for vision, text, and audio tasks that run on-device without sending input data to external servers.

Yes, if you need on-device ML inference for vision, text, or audio tasks and accept alpha-stage stability. The permissive Apache 2.0 license, active maintenance, and zero known vulnerabilities make it low-risk. Medium install friction (compiled wheels, multiple dependencies) is typical for ML packages. Verify that pre-trained models suit your use case and that you can manage the privacy implications of metrics collection to Google.

Install

mediapipe on PyPI

pip install mediapipe

uv add mediapipe

poetry add mediapipe

Installing mediapipe

Before you install

Medium install friction due to compiled wheels for multiple platforms (arm64, x86_64, Windows, macOS). Active maintenance with recent releases; repo shows 36617 stars and last commit on 2026-08-12. Seven runtime dependencies including numpy, opencv-contrib-python, and matplotlib add to setup complexity.

License in practice

Apache 2.0 permissive license allows commercial and private use with minimal restrictions. You must include a copy of the license and state significant changes, but can modify and distribute freely.

Quickstart

pip install mediapipe

import mediapipe as mp
from mediapipe.tasks import vision

# Load a pre-built model and run inference
detector = vision.ObjectDetector.create_from_model_path(model_path)

Requires a compatible Python version (3.9, 3.10, 3.11, or 3.12) and a pre-trained model file; on-device processing means no internet requirement but models must be available locally.

Verify before relying

• Specific model accuracy benchmarks or latency numbers for different tasks (vision, text, audio).
• Whether legacy solutions mentioned in docs are still functional or fully deprecated.
• Exact hardware requirements or GPU acceleration support details.
• Scope and frequency of metrics collection sent to Google servers.

Package facts

Evidence: mediapipe-1.0.1-py3-none-macosx_11_0_arm64.whl; mediapipe-1.0.1-py3-none-manylinux_2_28_aarch64.whl; mediapipe-1.0.1-py3-none-manylinux_2_28_x86_64.whl; mediapipe-1.0.1-py3-none-win_amd64.whl; mediapipe-1.0.1-py3-none-win_arm64.whl

Keywords: mediapipe

Tags

More Software Development packages

Further reading