Colorism in High Fashion (featuring: K-Means Clustering)
summary
Transforms twenty years of Vogue covers into ~250 per-cover datasets, fits a separate k-means clustering model per image (about 250 models), extracts average skin color, converts to HSL lightness, and shows only about two to four genuinely dark-skinned covers across two decades — a tokenism signal driven by Lupita Nyong'o’s repeated appearances. Pipeline uses OpenCV, remove.bg preprocessing, dozens of feature/hyperparameter variants and human review panels — a human-in-the-loop verification tool that chooses the final skin pixels.
AI-written summary · about
chapters (10)
- 0:00Colorism in high fashion
- 3:34Racism in fashion magazines
- 10:11K-means clustering for skin tones
- 13:22K-means on Vogue cover models
- 18:18Feature engineering for k-means
- 21:48Human-in-the-loop ML
- 23:13Colorism trends in Vogue
- 24:57Tokenism over time
- 27:24Beyond high fashion
- 30:25Q&A
related talks
42mAn Algorithm of Style
56mGlobal Bias in AI: When Western Data Shapes the World
27mClustering of high-content screen images to discover off-target phenotypes
42mModeling Aesthetic Identity- Building a Digital Twin
10mWhy diversity matters in Machine Learning communities
10mRecreating Photography of the 1850s in a Digital World
17mCalculating recommendations based on product images #bbuzz
25mComputer Vision in Melanoma Detection