Conell, Sarah Reiff Huang, Lingdong Levin, Golan Lincoln, Matthew National Gallery of Art InceptionV3 Features <div>On October 24-25, 2019 the National Gallery of Art in Washington, D.C.hosted a "data-thon" where multiple teams of art historians and data scientists worked with the museum's open collection data to study questions about the history and composition of the collections. <br></div><div><br></div><div>A joint team from Carnegie Mellon and the University of Pittsburgh used image features from a convolutional neural network to index the National Gallery of Art's images by visual similarity. This allowed the team to compare the visual distribution of different collections within the National Gallery, and with related parts of the Samuel H. Kress collection (distributed in museums around the country) as well as a portion of the Lessig Rosenwald collection (split between the National Gallery and the Library of Congress).</div><div><br></div><div>This deposit includes:</div><div><br></div><div>- a table of Inception V3 image features computed for images from the National Gallery of Art</div><div>- 2D visualizations of paintings, and of prints and drawings, clustered based on these computed features.<br></div><div>- Colorized visualizations showing the distribution of individual collectors' contributions in this visual similarity space</div><div>- Slides presented at the National Gallery</div><div><br></div><div>An interface for browsing these artworks by visual neighbors is available at https://dh-web.hss.cmu.edu/nga<br></div> computer vision;art;museum;National Gallery of Art, Washington D.C. 2019-11-06
    https://kilthub.cmu.edu/articles/figure/National_Gallery_of_Art_InceptionV3_Features/10061885
10.1184/R1/10061885.v1