Image Processing in the Browser
summary
Shows how to run TensorFlow.js models entirely in the browser using CPU, WebGL or WebGPU backends and Web Neural Network support, loading Coco SSD from npm and running detection on a video element at ~100 fps in demos. Builds a camera-based detector that flags a person and dog co-present for 5–10s to trigger alerts, with live examples returning class scores (laptop 72–99%, person ~80%). Notes WebGPU can be ~30x faster than CPU, models are retrainable on-device, and this approach preserves privacy and low latency.
AI-written summary · about
chapters (13)
- 0:00Introduction and speaker intro
- 3:38Why JavaScript for image processing
- 7:06Evolution of browser tech for ML
- 11:23TensorFlow and its platforms
- 14:17Defining the security camera problem
- 16:47Live coding demo setup
- 20:06Live coding detection results
- 22:38Recap of the live coding steps
- 23:41Real-world applications examples
- 25:08Built-in web AI overview
- 26:31Next steps and training models
- 27:57Benefits of client-side ML
- 28:40Resources and closing
related talks
41mTensorFlow.js 101: ML in the Browser and Beyond
20mTensorFlow.js Bringing Machine Learning to the Web and Beyond
54mDeep Learning in JS
29mSuperpowers for next gen web apps: Machine learning
26mMachine Learning and AI for Angular Developers
34mA Powerful Introduction to Machine Learning with Js
34mA Powerful Introduction to Machine Learning with Js
24mTensorFlow.js + Node