react-native-face-embeddings

A Node.js package for generating face embeddings using face-api.js and TensorFlow. Perfect for face recognition, comparison, and identity verification applications.

Installation

npm install react-native-face-embeddings

Prerequisites

This package requires the face-api.js model files. Download them from: https://github.com/vladmandic/face-api/tree/master/model

Place the following model files in a models directory in your project root:

• ssd_mobilenetv1_model-*
• face_landmark_68_model-*
• face_recognition_model-*

Usage

import { createEmbedding, findNearestMatch } from 'react-native-face-embeddings';

// Create embedding from a file path
const embedding1 = await createEmbedding('./photo1.jpg');

// Create embedding from base64 (React Native camera)
const embedding2 = await createEmbedding('data:image/jpeg;base64,...');

// Compare embeddings
const knownFaces = [
  { id: 'person1', embedding: embedding1 }
];

const match = findNearestMatch(knownFaces, embedding2);
if (match) {
  console.log(`Match found: ${match.id} (${match.confidence * 100}% confident)`);
}

API Reference

createEmbedding(imageInput)

Creates a 128-dimensional face embedding from an image.

Parameters:

• imageInput (string|Buffer|Uint8Array): Image source
  • File path: '/path/to/image.jpg'
  • File URI: 'file:///path/to/image.jpg'
  • Base64: 'data:image/jpeg;base64,...'
  • Buffer or Uint8Array

Returns: Promise<Array<number>> - 128-dimensional embedding

findNearestMatch(embeddings, targetEmbedding, threshold = 0.6)

Finds the best matching embedding from a list.

Parameters:

• embeddings (Array): Array of {id, embedding} objects
• targetEmbedding (Array): Embedding to match
• threshold (number): Maximum distance for a match (default: 0.6)

Returns: Object|null - {id, distance, confidence} or null

findAllMatches(embeddings, targetEmbedding, threshold = 0.6)

Finds all matching embeddings above threshold.

Returns: Array<Object> - Sorted array of matches

Example

See test/run.js for a complete example.

License

MIT © Harshadeep Donapati