|Google AI Recreates Lost Klimt Artworks|
|Written by David Conrad|
|Monday, 03 January 2022|
Too Good To Miss: In October we looked at a less obvious way that AI was influencing art. Can it really create lost works from what remains of it?
Google Arts and Culture has claimed another triumph for Machine Learning. After analyzing Gustav Klimt's use of color in extant artworks it has colorized the three giant canvases, Medicine, Jurisprudence and Philosophy destroyed during the Second World War.
The use of AI to colorize images is nothing new and in the past it has been criticized for using color that is inaccurate and either too subdued or too garish. In this exercise Google Arts and Culture, in collaboration with the Belvedure Museum in Vienna which has the world's largest collection of Gustav Klimt paintings, colorized lost artworks using the colors the artist actually used distilled from several sources into an algorithm.
The trio of paintings are known as "The Faculty Paintings" as they were originally commissioned in 1894 by the Austrian government as ceiling paintings for the assembly hall of the University of Vienna where the disciplines of Medicine, Jurisprudence and Philosophy were taught. In fact they never adorned that, or any, ceiling. Klimt began working on the paintings in 1898 and had to rent another studio to accommodate their enormous size - each over 13 feet tall. He first exhibited "Philosophy" in 1900 when the canvas met with fierce criticism. Numerous professors of the University of Vienna vehemently rejected the work. For them, Klimt's portrayal had nothing to do with the concept of philosophy as they understood it. In their eyes, Klimt had made a mockery of philosophy. An even greater scandal arose when "Medicine" was first displayed the following year.This time numerous politicians also intervened. and the Minister of Education was criticized for supporting Klimt.
The upshot was that Klimt cancelled the commission, repaying the large sum her had been paid.His patron, the Austrian industrialist Ledrerer acquired one of the works, Philosophy in 1905 and then bought Jurispudence in 1919. Both these painting were stolen by the Nazis sometime after 1938 and they ended up in Schloss Immendorf along with the third painting Medecine which, along with many other works by Klimt had been in the Belvedere. The day before the war ended, SS officers burnt the castle as the Nazis refused to have their art confiscated by the Russians and the painting were gone, with only black and white photographs and descriptions remaining.
As the video explains Dr. Franz Smola, a curator at Belvedere, sourced all the comments mentioning the Faculty Paintings and went on to match the scenes and color comments with Klimt’s remaining paintings. Using all this information, Emil Wallner, a resident at the Google Arts & Culture Lab, developed an algorithm to restore the Faculty Paintings. Instead of manually coloring the paintings, Wallner’s algorithm does a statistical analysis of Klimt’s existing artworks and learns how to mimic Klimt’s colorization style.
The model used has a similar structure to DeOldify developed by Jason Antic in 2018, see DeOldify - Auto Colorization and was trained on 91749 artworks from Google Arts & Culture to enable it to learn object boundaries, textures, and frequent compositions in artworks. As a final step, it was trained it on Klimt’s colored paintings to create a colorization bias towards color themes from Klimt’s artworks. While this doen't model Klimt’s artistic style in full, it gives "a prejudice for moods, colors, and recurring motifs in Klimt’s paintings". Also to enhance accuracy the algorithm used Dr. Smola’s research so where it is known that a certain object has a specific color, that color was added directly to the black and white photos and colors from Klimt's other works were used to develop the algorithm that replicated Klimt's style.
While we can't know for sure that the results are faithful to the lost originals, we now have something that art critics and art lovers alike can debate.
or email your comment to: email@example.com
|Last Updated ( Monday, 03 January 2022 )|