FaceAPI

Face detection and recognition libary for Browser and NodeJS implemented on top of TensorFlow/JS

Live Demo: https://vladmandic.github.io/face-api/demo/webcam.html

Documentation

• Tutorial
• API Specification

Note

This is updated face-api.js with latest available TensorFlow/JS as the original is not compatible with tfjs 2.0+.
Forked from face-api.js version 0.22.2 which was released on March 22nd, 2020

Currently based on TensorFlow/JS 3.3.0

Why? I needed Face-API that does not cause version conflict with newer versions of TensorFlow
And since original Face-API was open-source, I've released this version as well

Changes ended up being too large for a simple pull request
and it ended up being a full-fledged version on its own

Plus many features were added since original inception

Examples

Browser

Browser example that uses static images and showcases both models
as well as all of the extensions is included in /demo/index.html
Example can be accessed directly using Git pages using URL:
https://vladmandic.github.io/face-api/demo/index.html

Browser example that uses live webcam is included in /demo/webcam.html
Example can be accessed directly using Git pages using URL:
https://vladmandic.github.io/face-api/demo/webcam.html

Demo using FaceAPI to process images
Note: Photos shown below are taken by me

Demo using FaceAPI to process live webcam

NodeJS

Two NodeJS examples are:

• /demo/node-singleprocess.js:
  Regular usage of FaceAPI from NodeJS
• /demo/node-multiprocess.js:
  Multiprocessing showcase that uses pool of worker processes
  (node-multiprocess-worker.js)
  Main starts fixed pool of worker processes with each worker having
  it's instance of FaceAPI
  Workers communicate with main when they are ready and main dispaches
  job to each ready worker until job queue is empty

2021-03-14 08:42:03 INFO:  @vladmandic/face-api version 1.0.2
2021-03-14 08:42:03 INFO:  User: vlado Platform: linux Arch: x64 Node: v15.7.0
2021-03-14 08:42:03 INFO:  FaceAPI multi-process test
2021-03-14 08:42:03 STATE:  Main: started worker: 1888019
2021-03-14 08:42:03 STATE:  Main: started worker: 1888025
2021-03-14 08:42:04 STATE:  Worker: PID: 1888025 TensorFlow/JS 3.3.0 FaceAPI 1.0.2 Backend: tensorflow
2021-03-14 08:42:04 STATE:  Worker: PID: 1888019 TensorFlow/JS 3.3.0 FaceAPI 1.0.2 Backend: tensorflow
2021-03-14 08:42:04 STATE:  Main: dispatching to worker: 1888019
2021-03-14 08:42:04 STATE:  Main: dispatching to worker: 1888025
2021-03-14 08:42:04 DATA:  Worker received message: 1888019 { image: 'demo/sample1.jpg' }
2021-03-14 08:42:04 DATA:  Worker received message: 1888025 { image: 'demo/sample2.jpg' }
2021-03-14 08:42:06 DATA:  Main: worker finished: 1888025 detected faces: 3
2021-03-14 08:42:06 STATE:  Main: dispatching to worker: 1888025
2021-03-14 08:42:06 DATA:  Worker received message: 1888025 { image: 'demo/sample3.jpg' }
2021-03-14 08:42:06 DATA:  Main: worker finished: 1888019 detected faces: 3
2021-03-14 08:42:06 STATE:  Main: dispatching to worker: 1888019
2021-03-14 08:42:06 DATA:  Worker received message: 1888019 { image: 'demo/sample4.jpg' }
2021-03-14 08:42:07 DATA:  Main: worker finished: 1888025 detected faces: 3
2021-03-14 08:42:07 STATE:  Main: dispatching to worker: 1888025
2021-03-14 08:42:07 DATA:  Worker received message: 1888025 { image: 'demo/sample5.jpg' }
2021-03-14 08:42:08 DATA:  Main: worker finished: 1888019 detected faces: 4
2021-03-14 08:42:08 STATE:  Main: dispatching to worker: 1888019
2021-03-14 08:42:08 DATA:  Worker received message: 1888019 { image: 'demo/sample6.jpg' }
2021-03-14 08:42:09 DATA:  Main: worker finished: 1888025 detected faces: 5
2021-03-14 08:42:09 STATE:  Main: worker exit: 1888025 0
2021-03-14 08:42:09 DATA:  Main: worker finished: 1888019 detected faces: 4
2021-03-14 08:42:09 INFO:  Processed 15 images in 5944 ms
2021-03-14 08:42:09 STATE:  Main: worker exit: 1888019 0

Note that @tensorflow/tfjs-node or @tensorflow/tfjs-node-gpu
must be installed before using NodeJS example

Installation

Face-API ships with several pre-build versions of the library:

• dist/face-api.js: IIFE format for client-side Browser execution
  with TFJS pre-bundled
• dist/face-api.esm.js: ESM format for client-side Browser execution
  with TFJS pre-bundled
• dist/face-api.esm-nobundle.js: ESM format for client-side Browser execution
  without TFJS pre-bundled
• dist/face-api.node.js: CommonJS format for server-side NodeJS execution
  without TFJS pre-bundled
• dist/face-api.node-gpu.js: CommonJS format for server-side NodeJS execution
  without TFJS pre-bundled and optimized for CUDA GPU acceleration
• dist/face-api.node-cpu.js: CommonJS format for server-side NodeJS execution
  without TFJS pre-bundled and using JS engine for platforms where tensorflow binary library version is not available

Defaults are:

{
  "main": "dist/face-api.node-js",
  "module": "dist/face-api.esm.js",
  "browser": "dist/face-api.esm.js",
}

Bundled TFJS can be used directly via export: faceapi.tf

Reason for additional nobundle version is if you want to
include a specific version of TFJS and not rely on pre-packaged one

FaceAPI is compatible with TFJS 2.0+

All versions include sourcemap

There are several ways to use Face-API:

1. IIFE script

Recommened for quick tests and backward compatibility with older Browsers that do not support ESM such as IE

This is simplest way for usage within Browser
Simply download dist/face-api.js, include it in your HTML file & it's ready to use

<script src="dist/face-api.js"><script>

IIFE script bundles TFJS and auto-registers global namespace faceapi within Window object which can be accessed directly from a <script> tag or from your JS file.

2. ESM module

Recommended for usage within Browser

2.1. Direct Import

To use ESM import directly in a Browser, you must import your script (e.g. index.js) with a type="module"

  <script src="./index.js" type="module">

and then in your index.js

  import * as faceapi from 'dist/face-api.esm.js';

2.2. With Bundler

Same as above, but expectation is that you've installed @vladmandic/faceapi package:

  npm install @vladmandic/face-api 

and that you'll package your application using a bundler such as webpack, rollup or esbuild
in which case, you do not need to import a script as module - that depends on your bundler configuration

  import * as faceapi from '@vladmandic/face-api';

or if your bundler doesn't recognize recommended type, force usage with:

  import * as faceapi from '@vladmandic/face-api/dist/face-api.esm.js';

or to use non-bundled version

  import * as tf from `@tensorflow/tfjs`;
  import * as faceapi from '@vladmandic/face-api/dist/face-api.esm-nobundle.js';

3. NPM module

3.1. Import CommonJS

Recommended for NodeJS projects

Node: Face-API for NodeJS does not bundle TFJS due to binary dependencies that are installed during TFJS installation

Install with:

  npm install @tensorflow/tfjs-node
  npm install @vladmandic/face-api 

And then use with:

  const tf = require('@tensorflow/tfjs-node')
  const faceapi = require('@vladmandic/face-api');

If you want to force CommonJS module instead of relying on recommended field:

  const faceapi = require('@vladmandic/face-api/dist/face-api.node.js');

If you want to GPU Accelerated execution in NodeJS, you must have CUDA libraries already installed and working
Then install appropriate version of Face-API:

  npm install @tensorflow/tfjs-node
  npm install @vladmandic/face-api 

And then use with:

  const tf = require('@tensorflow/tfjs-node-gpu')
  const faceapi = require('@vladmandic/face-api/dist/face-api.node-gpu.js'); // this loads face-api version with correct bindings for tfjs-node-gpu

If you want to use Face-API in a NodeJS on platforms where NodeJS binary libraries are not supported, you can use JavaScript CPU backend.

  npm install @tensorflow/tfjs
  npm install @vladmandic/face-api 

And then use with:

  const tf = require('@tensorflow/tfjs')
  const faceapi = require('@vladmandic/face-api/dist/face-api.node-cpu.js');

If you want to use graphical functions inside NodeJS,
you must provide appropriate graphical library as
NodeJS does not include implementation for DOM elements
such as HTMLImageElement or HTMLCanvasElement:

Install Canvas for NodeJS:

npm install canvas

Patch NodeJS environment to use newly installed Canvas library:

const canvas = require('canvas');
const faceapi = require('@vladmandic/face-api');

const { Canvas, Image, ImageData } = canvas
faceapi.env.monkeyPatch({ Canvas, Image, ImageData })

Weights

Pretrained models and their weights are includes in ./model.

Test & Dev Web Server

Built-in test&dev web server can be started using

npm run dev

By default it starts HTTP server on port 8000 and HTTPS server on port 8001 and can be accessed as:

• https://localhost:8001/demo/index.html
• https://localhost:8001/demo/webcam.html

2021-03-14 08:41:09 INFO:  @vladmandic/face-api version 1.0.2
2021-03-14 08:41:09 INFO:  User: vlado Platform: linux Arch: x64 Node: v15.7.0
2021-03-14 08:41:09 INFO:  Build: file startup all target: es2018
2021-03-14 08:41:09 STATE:  HTTP server listening: 8000
2021-03-14 08:41:09 STATE:  HTTP2 server listening: 8001
2021-03-14 08:41:09 STATE:  Monitoring: [ 'package.json', 'demo', 'src', [length]: 3 ]
2021-03-14 08:41:10 STATE:  Build for: browserBundle type: tfjs: { modules: 1258, moduleBytes: 4040087, imports: 7, importBytes: 276, outputBytes: 1072314, outputFiles: 'dist/tfjs.esm.js' }
2021-03-14 08:41:10 STATE:  Build for: browserBundle type: iife: { imports: 160, importBytes: 1305679, outputBytes: 1151683, outputFiles: 'dist/face-api.js' }
2021-03-14 08:41:10 STATE:  Build for: browserBundle type: esm: { imports: 160, importBytes: 1305679, outputBytes: 1151520, outputFiles: 'dist/face-api.esm.js' }

Build

If you want to do a full rebuild, either download npm module

npm install @vladmandic/face-api
cd node_modules/@vladmandic/face-api

or clone a git project

git clone https://github.com/vladmandic/face-api
cd face-api

Then install all dependencies and run rebuild:

npm install
npm run build

Build process uses script build.js that creates optimized build for each target:

> @vladmandic/[email protected] build
> rimraf dist/* types/* typedoc/* && node server/build.js

2021-03-14 08:39:21 INFO:  @vladmandic/face-api version 1.0.2
2021-03-14 08:39:21 INFO:  User: vlado Platform: linux Arch: x64 Node: v15.7.0
2021-03-14 08:39:21 INFO:  Build: file startup all target: es2018
2021-03-14 08:39:21 STATE:  Build for: node type: tfjs: { imports: 1, importBytes: 143, outputBytes: 731, outputFiles: 'dist/tfjs.esm.js' }
2021-03-14 08:39:21 STATE:  Build for: node type: node: { imports: 160, importBytes: 234096, outputBytes: 85371, outputFiles: 'dist/face-api.node.js' }
2021-03-14 08:39:21 STATE:  Build for: nodeGPU type: tfjs: { imports: 1, importBytes: 147, outputBytes: 735, outputFiles: 'dist/tfjs.esm.js' }
2021-03-14 08:39:21 STATE:  Build for: nodeGPU type: node: { imports: 160, importBytes: 234100, outputBytes: 85379, outputFiles: 'dist/face-api.node-gpu.js' }
2021-03-14 08:39:21 STATE:  Build for: nodeCPU type: tfjs: { imports: 1, importBytes: 138, outputBytes: 726, outputFiles: 'dist/tfjs.esm.js' }
2021-03-14 08:39:21 STATE:  Build for: nodeCPU type: node: { imports: 160, importBytes: 234091, outputBytes: 85370, outputFiles: 'dist/face-api.node-cpu.js' }
2021-03-14 08:39:21 STATE:  Build for: browserNoBundle type: tfjs: { imports: 1, importBytes: 276, outputBytes: 244, outputFiles: 'dist/tfjs.esm.js' }
2021-03-14 08:39:21 STATE:  Build for: browserNoBundle type: esm: { imports: 160, importBytes: 233609, outputBytes: 82634, outputFiles: 'dist/face-api.esm-nobundle.js' }
2021-03-14 08:39:22 STATE:  Build for: browserBundle type: tfjs: { modules: 1258, moduleBytes: 4040087, imports: 7, importBytes: 276, outputBytes: 1072314, outputFiles: 'dist/tfjs.esm.js' }
2021-03-14 08:39:22 STATE:  Build for: browserBundle type: iife: { imports: 160, importBytes: 1305679, outputBytes: 1151683, outputFiles: 'dist/face-api.js' }
2021-03-14 08:39:22 STATE:  Build for: browserBundle type: esm: { imports: 160, importBytes: 1305679, outputBytes: 1151520, outputFiles: 'dist/face-api.esm.js' }
2021-03-14 08:39:22 INFO:  Compile typings: [ 'src/index.ts', [length]: 1 ]
2021-03-14 08:39:27 INFO:  Update Change log: [ '/home/vlado/dev/face-api/CHANGELOG.md', [length]: 1 ]
2021-03-14 08:39:27 INFO:  Generate TypeDocs: [ 'src/index.ts', [length]: 1 ]

Face Mesh

FaceAPI landmark model returns 68-point face mesh as detailed in the image below:

Differences

Compared to face-api.js version 0.22.2:

• Compatible with TensorFlow/JS 2.0+ & 3.0+
• Compatible with WebGL, CPU and WASM TFJS Browser backends
• Compatible with both tfjs-node and tfjs-node-gpu TFJS NodeJS backends
• Updated all type castings for TypeScript type checking to TypeScript 4.2
• Switched bundling from UMD to ESM + CommonJS with fallback to IIFE Resulting code is optimized per-platform instead of being universal Fully tree shakable when imported as an ESM module Browser bundle process uses ESBuild instead of Rollup
• Typescript build process now targets ES2018 and instead of dual ES5/ES6 Resulting code is clean ES2018 JavaScript without polyfills
• Removed old tests, docs, examples
• Removed old package dependencies (karma, jasmine, babel, etc.)
• Updated all package dependencies
• Updated TensorFlow/JS dependencies since backends were removed from @tensorflow/tfjs-core
• Updated mobileNetv1 model due to batchNorm() dependency
• Added version class that returns JSON object with version of FaceAPI as well as linked TFJS
• Added test/dev built-in HTTP & HTTPS Web server
• Removed mtcnn and tinyYolov2 models as they were non-functional in latest public version of Face-API Which means valid models are tinyFaceDetector and mobileNetv1 If there is a demand, I can re-implement them back.
• Added face angle calculations that returns roll, yaw and pitch
• Added typdoc automatic API specification generation during build
• Added changelog automatic generation during build

Credits

• Original project: Face-API
• Original model weighs: Face-API
• ML API Documentation: Tensorflow/JS