Hand3000

༼ つ ◕_◕ ༽つ

Currently an experimental and rudimentary decorator for TensorFlow's Handpose.

Demo

What

TensorFlow's handpose only outputs a set of points described by x, y, z coordinates. This little experimental "decorator" just evaluates these points and decorates the output of estimateHands with values such as distance, spread, roll, pitch, isFist. If you have any additions, suggestions, please feel free to make a PR!

• [ ] isPointing
• [ ] Some other inference?

| Value | Description | |------------|-------------------------------------------------------------------------------------------------------------------------| | distance | An estimate of the hand distance from the camera. Basically a measurement between 2 of the "static" points in the model | | spread | An average calculation of the distance between fingertips | | roll | Like how an airplane moves, roll is the rotation in the x-axis | | pitch | Like how an airplane moves, pitch is the rotation in the z-axis | | isFist | Not glamorously named, but it is what it says. A rough estimation IF a user's hand is closed | | isPointing | Coming soon |

Usage

yarn add hand3000 or npm i hand3000

  import * as handpose from '@tensorflow-models/handpose';
  import DecorateHand from 'hand3000';

  // Initialize tensorflow as usual
  const model = await handpose.load();

  // Feed the model into Hand3000
  DecorateHand(model);

  // This just tweaks the estimateHand methods and adds
  // above computed values in the return of estimateHands
  const predictions = await model.estimateHands(source);