PixPlot is open-source software being developed within the Yale Digital Humanities Lab (DHLab). It uses a convolutional neural network and Uniform Manifold Approximation and Projection (UMAP) in order to cluster visually similar images near one another in a web browser.

As a collaborator on the project, I advocate for users, contributing design and functionality ideas that improve the user experience (with users including PixPlot creators and viewers), as well as the legibility of the visualization. I also curate sample datasets, write copy for project demos, and conduct testing on newly implemented features.

For the source code—written by software developer Douglas Duhaime—visit the DHLab’s GitHub repository. Additional team members include Peter Leonard (Director) and Monica Ong Reed (User Experience Designer).

PixPlots I created for testing purposes

MHL PixPlot (24,995 images, PixPlot version 0.098)

Images are from the Medical Heritage Library’s Flickr account, which I accessed through Flickr’s API.

SI PixPlot (13,391 images, version 0.094)

Images are from the Smithsonian Open Access API.

Bain News Service PixPlot (24,026 images, version 0.058)

Images are from the Library of Congress’s Flickr account, which I accessed through Flickr’s API.