What is “style” for computers?
The Substance of Style: Connoisseurship Between Art History and Digital Technologies”
Author: Alison Langmead, Christopher J. Nygren, University of Pittsburgh
One of the major questions facing the field of digital art history is: how might computers help us understand the ongoing disciplinary valence of the concept of artistic style? To get at this critically important issue, we believe the first step is to ask whether or not a computational investigation into the visual qualities of works of art and architecture can successfully identify a “style.” Our own reflections on this concept are based on a long-term collaboration in digital art history which sought to investigate the connoisseurial methods of Giovanni Morelli using computational techniques. These attempts reproduce Morelli’s method computationally forced us to confront, in a stark and sometimes startling way, the fact that art historians appear fundamentally reticent to assert “where” or “in what” such a thing as “style” resides. Is it embedded in brush strokes? The contour of a line? The repetition of a pattern? The extent to which computers will be helpful in the process of identifying style hinges on how art historians answer these kinds of questions. If visual forms are indeed a dominant marker of style, computers may yield essential insights that humans are unlikely to uncover on their own. If the meaningful markers of artistic style reside elsewhere, however, the role of computer vision in this research agenda will need to be re-theorized. We believe that the question of how style is conceived is crucial to thinking about the future of digital art history.
Dr. Alison Langmead holds a joint faculty appointment at the University of Pittsburgh between the Dietrich School of Arts and Sciences and the School of Computing and Information. She directs the Visual Media Workshop (VMW), a humanities lab located in the Department of the History of Art and Architecture that investigates material and visual culture in an environment that encourages technological experimentation. Alison also leads the Executive Committee overseeing Pitt’s graduate and undergraduate Digital Studies and Methods curricula, and serves as the principal contact for the DHRX: Digital Humanities Research initiative at Pitt.
Christopher J. Nygren
Dr. Christopher Nygren is associate professor of early modern art in the Department of the History of Art and Architecture at the University of Pittsburgh. His 2020 book, Titian’s Icons: Charisma, Tradition, and Devotion in the Italian Renaissance, published by Penn State University Press, re-examined one of the leading lights of Italian Renaissance painting to reveal the lasting impact of Christian icons on Titian’s career. Between 2017 and 2019 he served as PI for The Morelli Machine, an NSF funded project that tested the hypothesis that the nexus of style and authorship can be interrogated computationally.
Content vs. Style: Two sides of the same coin
Author: Peter Bell, Dirk Suckow, Prathmesh Madhu, Ronak Kosti, University of Erlangen-Nuremberg
A style can be interpreted as a genre, a specific technique, an individual signature of an artist or it can even point to the sense of general appearance, essence, nature or related, e.g., to an era, region/city, culture, nation. In computational perspective, style could refer to the data distribution across different domains. In this seminar, with the help of deep neural networks and generative adversarial networks, we discuss the following: a) Is the main content of an image enough for scene understanding, or does the style matter? b) Can we include the styles computationally to have a more informed understanding of how the models work? c) Will the style inclusion influence the performance of the model? d) If multiple styles are encoded inside a single image, will the behaviour of the network be explainable? With the focus on recognizing the characters in Art History, we discuss the variety of styles present in a single iconography like "Annunciation of the Lord", and how these styles can affect recognition of the individual characters. We follow it up with a special study on a public dataset called ArtDL - which depicts scenes or characters of the Christian art. Our approach should not only help to further develop methods of CV by including a broad spectrum of art-historical images, but vice versa to ask anew from a double perspective of "Bildwissenschaft" what constitutes style on an abstract level, as a predefined set of respective characteristics and in the sense of general appearance, essence, nature or manner.
Prof. Dr. Peter Bell - Digital Art Historian. 2015-2016 Managing Director of the Hartware MedienKunstVerein at Dortmunder U. 2017 Research Associate at the prometheus-Bildarchiv at the University of Cologne. Since November 2017 assistant professor of Digital Humanities with a focus on art history and affiliated to the Institute of Art History.
Dirk Suckow, M.A. - Art historian/Historian. Since October 2019 research associate at FAU Erlangen-Nürnberg, currently involved in the interdisciplinary project ‘Iconographics. Computational Understanding of Iconography and Narration in Visual Cultural Heritage’.
Prathmesh Madhu, M.Sc. - Since December 2018 pursuing PhD as DAAD Scholar in Digital Humanities and Medical Imaging at Pattern Recognition Lab, Computer Science department of FAU Erlangen-Nürnberg, Germany.
Dr. Ronak Kosti - Computer Scientist. Currently involved in a DFG-SSP project: ‘Image Synthesis as an Epistemic Method Towards Understanding Art’ at FAU Erlangen-Nürnberg. Prior to this project, he was working on a project called 'ICONOGRAPHICS' with focus on computational understanding of paintings and artworks.