BlazeFace hello world

BlazeFace is a neural network model that detects faces in images. It’s designed to be fast, to run at 30fps on mobile GPUs. There is a TensorFlow.js library for BlazeFace, which downloads the model, runs it in WebGL using TensorFlow.js, and wraps the raw model input/output with a friendly, semantic API. to start a demo, which captures and displays your webcam, runs BlazeFace against frames as often as possible, and draws the detected face landmarks on top of your webcam stream:

Here’s what I get when I run it against my own face:

Basic usage of this API is basically one function call; a pure function from input image to predicted faces:

<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@2.4"></script>

<script src="https://cdn.jsdelivr.net/npm/@tensorflow-models/blazeface@0.0.5"></script>

const model = await blazeface.load();
const predictions = await model.estimateFaces(webcamVideoEl, false);
console.log(predictions);

This will log an object looking something like:

[
  { // One detected face
    topLeft: [162.84341430664062,153.98446655273438],  // [x,y] coordinates
    bottomRight: [422.3966979980469,348.6485900878906],
    landmarks:[
      [238.96787643432617,204.8737621307373], // right eye
      [328.2145833969116,205.4714870452881],  // left eye
      [273.6716037988663,252.84512042999268], // nose
      [280.01041051000357,293.8540989160538], // mouth
      [206.6525173187256,226.03596210479736], // right ear
      [386.57989501953125,226.02698922157288] // left ear
    ],
    probability: [0.9997807145118713]  // Always a one-element array; a bit odd
  }
]

The topLeft and bottomRight coordinates define a “bounding box”, but I don’t know exactly what it’s supposed to bound. Certainly, it seems to always contain the six detected landmarks, but not precisely. My eventual goal is to draw a boundary around the head; the default bounding box is not necessarily helpful for this.

Under certain conditions, BlazeFace consistently recognized a face in my forehead, and was extremely confident about it:

Probably one of my weirder debugging sessions pic.twitter.com/AubMIkM1kI
— Jim Fisher (@MrJamesFisher) September 20, 2020

The bug seems to only happen when I use my high-resolution webcam feed. BlazeFace performs much more reliably with a lower-resolution webcam feed. This is very strange, because I believe the library resizes the input to 128×128 pixels before analyzing it. I’ll do a future post on the internals of this library, and how to use TensorFlow.js. This should help understand the weird forehead bug.

Tagged #programming, #web, #tensorflow, #machine-learning.

More by Jim

What does the dot do in JavaScript?

foo.bar, foo.bar(), or foo.bar = baz - what do they mean? A deep dive into prototypical inheritance and getters/setters. 2020-11-01

Smear phishing: a new Android vulnerability

Trick Android to display an SMS as coming from any contact. Convincing phishing vuln, but still unpatched. 2020-08-06

A probabilistic pub quiz for nerds

A “true or false” quiz where you respond with your confidence level, and the optimal strategy is to report your true belief. 2020-04-26

Time is running out to catch COVID-19

Simulation shows it’s rational to deliberately infect yourself with COVID-19 early on to get treatment, but after healthcare capacity is exceeded, it’s better to avoid infection. Includes interactive parameters and visualizations. 2020-03-14

The inception bar: a new phishing method

A new phishing technique that displays a fake URL bar in Chrome for mobile. A key innovation is the “scroll jail” that traps the user in a fake browser. 2019-04-27

The hacker hype cycle

I got started with simple web development, but because enamored with increasingly esoteric programming concepts, leading to a “trough of hipster technologies” before returning to more productive work. 2019-03-23

Project C-43: the lost origins of asymmetric crypto

Bob invents asymmetric cryptography by playing loud white noise to obscure Alice’s message, which he can cancel out but an eavesdropper cannot. This idea, published in 1944 by Walter Koenig Jr., is the forgotten origin of asymmetric crypto. 2019-02-16

How Hacker News stays interesting

Hacker News buried my post on conspiracy theories in my family due to overheated discussion, not censorship. Moderation keeps the site focused on interesting technical content. 2019-01-26

My parents are Flat-Earthers

For decades, my parents have been working up to Flat-Earther beliefs. From Egyptology to Jehovah’s Witnesses to theories that human built the Moon billions of years in the future. Surprisingly, it doesn’t affect their successful lives very much. For me, it’s a fun family pastime. 2019-01-20

The dots do matter: how to scam a Gmail user

Gmail’s “dots don’t matter” feature lets scammers create an account on, say, Netflix, with your email address but different dots. Results in convincing phishing emails. 2018-04-07

The sorry state of OpenSSL usability

OpenSSL’s inadequate documentation, confusing key formats, and deprecated interfaces make it difficult to use, despite its importance. 2017-12-02

I hate telephones

I hate telephones. Some rational reasons: lack of authentication, no spam filtering, forced synchronous communication. But also just a visceral fear. 2017-11-08

The Three Ts of Time, Thought and Typing: measuring cost on the web

Businesses often tout “free” services, but the real costs come in terms of time, thought, and typing required from users. Reducing these “Three Ts” is key to improving sign-up flows and increasing conversions. 2017-10-26

Granddad died today

Granddad died. The unspoken practice of death-by-dehydration in the NHS. The Liverpool Care Pathway. Assisted dying in the UK. The importance of planning in end-of-life care. 2017-05-19

How do I call a program in C, setting up standard pipes?

A C function to create a new process, set up its standard input/output/error pipes, and return a struct containing the process ID and pipe file descriptors. 2017-02-17

Your syntax highlighter is wrong

Syntax highlighters make value judgments about code. Most highlighters judge that comments are cruft, and try to hide them. Most diff viewers judge that code deletions are bad. 2014-05-11

Want to build a fantastic product using LLMs? I work at Granola where we're building the future IDE for knowledge work. Come and work with us! Read more or get in touch!

This page copyright James Fisher 2020. Content is not associated with my employer. Found an error? Edit this page.

BlazeFace hello world

Similar posts

More by Jim