How to Cheat at Life with JS, CV, and AR
summary
Builds a phone browser app that uses JavaScript, canvas imageData, thresholding, contour tracing (more neighborhood tracing), perspective transform and white‑point correction to extract cards then classifies hue, count by contour area, shading by inner luminosity and shape by convex hull and area. Solves sets by brute force — 220 possible sets checked in microseconds — yielding about 300,000 sets per minute. Initial model overfits to a few images, retrains with many captures (poor man's machine learning), and sometimes beats Susan but still fails under tricky lighting.
AI-written summary · about
chapters (9)
- 0:00Intro
- 5:10The Problem
- 7:31The Image Data Object
- 8:23Neighborhood Tracing
- 11:11Perspective Transform
- 16:12Set Finding
- 17:40Overfitting
- 20:28Gameplay
- 23:27Other Examples
related talks
32mMagic of Computer Vision Unraveled
39mWhose turn is it anyway? Augmented reality board games.
16mImage Processing in JavaScript
32mSelf-Portraits in JavaScript
26mSelf-Portraits in JavaScript
36mMaking a Robot Eye with JavaScript
32mTeach JS About Aesthetics with Machine Learning
37mJavaScript Machine Learning from Magic to Science