TensorFlow.js

on YouTube ↗48m

" summary

TensorFlow.js runs real training and inference inside browsers using WebGL and WebAssembly backends and a layers API compatible with Keras. Demo loads a converted MobileNet (≈18 MB split into 4 MB shards) over HTTP, runs webcam inference via WebRTC and canvas at interactive rates; same JavaScript code takes ~3.3 s on plain JS CPU, ~49 ms on integrated graphics, ~5 ms on a GTX1080, and ~3 ms on server CUDA. Browser can also train small datasets (Iris, 40 epochs) and save models locally.

AI-written summary · about

" chapters (8)

1. 0:00Introduction to TensorFlow.js
2. 3:50Why TensorFlow.js matters
3. 10:13Performance and hardware acceleration
4. 15:57Basic TensorFlow.js app structure
5. 19:50Demo with Iris dataset
6. 27:45WebRTC and real-time ML
7. 33:27Live camera demo walkthrough
8. 46:13Future possibilities and resources

" related talks

• 20mTensorFlow.js Bringing Machine Learning to the Web and BeyondNick Kreeger, Nikhil Thorat · JSConf EU 2019
• 24mTensorFlow.js + NodeNick Kreeger, Yannick Assogba · EmpireJS 2018
• 31mIntroduction to Machine Learning in Tensorflow.jsNick Kreeger · ForwardJS 2019
• 41mTensorFlow.js 101: ML in the Browser and BeyondJason Mayes · ML Conf EU 2020
• 37mTensorFlow.js - Bringing ML and Linear Algebra to Node.jsSandeep Gupta, Ping Yu · Node+JS Interactive 2019
• 21mA Journey into the World of Machine Learning with TensorFlow.jsAaron Ma · ng-conf 2020: Hardwired
• 01h 04mTensorFlow in your BrowserOswald Campesato · Silicon Valley Code Camp 2018
• 28mMachine Learning on the web with TensorFlow.jsJason Mayes, Sandeep Gupta, Ping Yu · Google I/O 2021