font.fish is a browser-based tool for exploring thousands of fonts. My Yale undergraduate thesis advised by Holly Rushmeier.
font.fish uses Keras’s Inception v3 bindings to generate a featurization space in 2,048 dimensions. Two dimensionality reduction techniques, Uniform Manifold Approximation and Projection (UMAP) and t-distributed Stochastic Neighbor Embedding (t-SNE), reduce the feature space to 2 dimensions. The 2D space is visualized using Three.js.
Inspired by the Yale Digital Humanities Lab’s PixPlot.