Prejudice against people of color with a darker skin tone (called “colorism”) is prevalent in high fashion, but because it is subtler than overt racism it is harder to quantify. This talk will go into the technical details of transforming twenty years of Vogue cover photos into a numeric dataset that measures skin tone lightness, which, when visualized, showcases the striking lack of diversity on the cover of Vogue magazine.

This spring, I worked with The Pudding to explore diversity in fashion through the lens of colorism. We applied quantitative methods to an area where research is overwhelmingly qualitative by developing a metric that measures how light the skin of a Vogue model looks in a photograph.

The result was a case study that went pretty viral. The article itself skimmed over the techniques that were used to gather data, so I will speak about the intricacies of building this dataset:

• Interpreting a photograph as a three-dimensional array of floats
• Using HSL color representation to extract meaningful features from a photograph
• Fitting and tuning k-means models to identify skin in a photograph
• Using human-in-the-loop machine learning to improve model results
• Visualizing color data

This talk will be accessible to people with a beginner level knowledge of machine learning, and no background in fashion.