Challenges and Trade-offs for In-Browser Computer Vision Applications
summary
The talk examines practical limits of running computer vision inside web browsers, focusing on document/A4 detection that yields many false positives on laminated or low-contrast photos. It covers trade-offs between accuracy and speed (many passes per frame, hyperparameter tuning), difficulties optimizing morphological operations and OCR for browser CPU/GPU and memory, and performance/network issues when handling very large images (10–50 MB), plus engineering workarounds used to make such apps viable.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 14:00In-browser CV promise
- 28:00Classical CV challenges
- 42:00Performance vs UX
- 56:00Large image data
- 1:03:19Q&A
related talks
30mIntroduction to Computer Vision with OpenCV
32mMagic of Computer Vision Unraveled
29mImage Processing in the Browser
41mEscaping the Cloud: High-Performance AI in your Browser
24mWeb Apps of the Future With Web AI
26mSelf-Portraits in JavaScript
16mImage Processing in JavaScript
04mBrowser Agents Don't Need Better Models. They Need Better Eyes.