Self-Portraits in JavaScript

on YouTube ↗32m

" summary

A browser app captures webcam frames, freezes them to canvas, and either sends the image blob to cloud endpoints that return rich landmark and attribute JSON or processes them locally with pre-trained models; using TensorFlow.js & face-api.js requires downloading models and taxes the CPU but runs offline, while calling AWS Google Cloud & Azure returns detailed age/gender/smile confidences instantly. The native Shape Detection API is fastest but experimental and buggy behind flags. Neural style transfer via magenta can paint convincing results. The tool works with occasional landmark errors and clear privacy tradeoffs.

AI-written summary · about

" chapters (6)

1. 0:00Start
2. 10:36Web-based Services
3. 11:27Shape Detection API (Native)
4. 14:39Azure Face API Google Cloud Vision
5. 20:41Pure Client-side Approaches
6. 30:14Hong Kong Protests

" related talks

• 49mReal Time Face Detection through a Webcam Using AIObaid Sarvana · Data Con LA 2019
• 39mAI + JavaScript RocksAsim Hussain · FrontEnd Con 2018
• 51mRealTime Exoskeleton-Mapping and Body Segmentation on the Browser with Tensorflow.jsItalo Jose · CityJS Conference 2020
• 31mNon-traditional usage of Cloud for Science ArtDmitry Soshnikov · Infobip Shift 2021
• 48mThe Future of Machine Learning & JavaScriptAsim Hussain · JS Fest 2019 Autumn
• 11mAI-powered JavaScript applicationsAsim Hussain · The Lead Developer London 2019
• 54mDeep Learning in JSAshi Krishnan · CSS-Minsk-JS 2018
• 35mHow to AI in JS?Asim Hussain · Angular Connect 2018