tanırım
AI & Model

MediaPipe Hand Landmarker

ModelPose EstimationBrowser

A ready-to-use model bundle from Google's MediaPipe family for tracking hands in real time. It works in two stages. First a palm detection model locates the hand region, then a landmark model extracts 21 joint points per hand within it. It can track two hands at once. It runs in the browser with WebGL or WASM, directly on the user's device, with no server.

General Properties

Architecture / Approach
Two-stage: palm detection + 21-point hand landmark model
Trained On
~30K real images + synthetic hand models
Model Size
~7.8 MB
Input Resolution
192×192 (palm) + 224×224 (landmark)
License
Apache-2.0
// Help

Frequently Asked Questions

It is used to track hands and finger joints in the camera feed in real time. Its speed on different devices was measured in the Lab's in-browser benchmark.

The model runs on the graphics card through WebGL, falling back to the processor with WASM when WebGL is unavailable. Either way, your footage never leaves your device.

Apache-2.0. The model file and runtime are free to use.