This repository contains several analyses and samples from a hair dataset. Important features:
Hair_Maps contains the hair "heat maps" for every image in the yearbook dataset created by Ginosar et al. The maps are probabilistic (the value of every matrix element is in the interval 0-1).
Looks_by_Decades contains a folder of sample images representing every decade from 1930-2010. An accompanying .csv file provides demographic info (male/female label, decade) about each picture.
Size_by_Year contains several analysis scripts to
-
Measure hair density in every picture in the dataset,
- The script 1_Measure_Hair_Density.R puts out hair_f_density.csv and hair_m_density.csv, providing the hair density in all the male and all the female yearbook photos.
-
Analyze the median hair size from each year, and
- The script 2_Analyze_Size_Trends.R takes the .csv files from the first step and generates trend_f_smoothed.csv and trend_m_smoothed.csv, which give the median hair size from each year as well as the smoothed time series.
-
Estimate the hair size in the sample images in the "Looks_by_Decades" directory.
- The script 3_Size_in_Sample_Images.R estimates the hair density in every image selected as representative of a decade (in Looks_by_Decades) and outputs the estimates in Sample_image_hair_size.csv.
Pics_from_hairspace contains scripts to train and sample a variational autoencoder, which generates pictures of hair along four principal axes. Trained models are not included as these files are quite large.
Mullets_n_Beehives contain labeled examples of mullets, beehives, and straighthair, along with a tabular data regarding the frequency of each look per year.
Gender_trends contains a script to estimate the disriminability of two binary labels (male/female) in the dataset (Discriminability_by_Year.R) using random forest classification. The input is coordinates.csv, which contains the four features measured for each hair map, and the output is Discriminability_by_year.csv. The output file includes both actual discriminability-per-year as well as a smoothed timeseries.