Teach JS About Aesthetics with Machine Learning
summary
Claims a practical recipe for judging photos: extract numeric features (rule of thirds, loneliness, simplicity, contrast, color theory), reduce correlation with wavelets, then train classifiers rather than pontificate about aesthetics. Reports surprising figures: a paper uses 56 features and reaches 70% accuracy, mentions $3.2 billion as context. Demonstrations use pixel-based nets and libraries like brain.js and conet Js comnet Js and emphasize which features and tradeoffs actually move the needle.
AI-written summary · about
chapters (4)
related talks
16mPerceptual Evaluations: Evals for Aesthetics
32mMagic of Computer Vision Unraveled
22mKeynote talk - The future of machine learning & JS
39mAI + JavaScript Rocks
36mPractical Machine Learning for Front End Developers
36mMaking a Robot Eye with JavaScript
35mHow to AI in JS?
34mA Powerful Introduction to Machine Learning with Js