Page 287 - The Black Box Book

W H AT T HE M A C HI NE The final section Curator of Symbols analyses iconographic anal-
VI S I O N C AN S E E ? ysis and genre classification. In this case, the neural network algorithm
has been trained to classify motifs and themes, attempting to capture
If we examine a painting in a gallery, we can see it contains their significance, identify the works on which they appear, and conduct
an array of different things. When we are looking at it from a distance, we frequency analysis to determine why their popularity reigns over time.
are likely to be drawn to its dimensions or overall composition. When we
edge closer, we discern the abstract style of the painting, our eyes follow D A T A B A S E
the lines, surfaces, edges and shapes. When the style of the painting is
representative, we start to explore a motif, a figure or the elements and Despite the fact that the principles of computer vision are rooted in linear
how they are spatially arranged. In the case of a portrayal of a human algebra, the data itself is more than merely mathematics. Arguably, it is
being, our eyes will naturally turn to the face, the features and gestures the responsibility of the curator or art historian to continuously use their
depicted, the position of the body or the attributes of the figure. We might critical eye concerning the specific data, take an active role in its selection
also be intrigued by the creative style, the schematic aspect of the image, and to also suggest changes to its composition, form, and processing
or the specifics that emerge from the artist’s handwork, such as brush- techniques. Here, the composition of sample datasets for machine learn-
strokes, contrasts, or the color scheme; we might even pay attention to ing can influence how the algorithm comes to evaluate and potentially
the unique and defining artistic signature. shape the outcome of the research. We will therefore proceed to outline
Machine Vision tools and human eyes both possess numerous the data that the Digital Curator handles.
ways and methods of visioning the world. Just as there are many ways The Digital Curator database can be compared to the collection
of observing with the human eye, there are a variety of machine vision of an art museum or gallery. More specifically, it is a collection of dig-
tools that can be used in the analysis. Each of these methods highlights ital reproductions of artworks that an algorithm has at its disposal to
a unique aspect of images as well as an interpretation. Thus, by the use search, filter, sort, label, group, and in orderlink, to create datasets and
of specialized algorithms, one can analyze in a different manner colors, shape their context. The works included in the database are accessible
contrasts, shapes, search for similar images in large collections, compare and processable by the program; the others, although they can be pub-
painting the handstyles of different artists, and classify artistic styles. lished online, are left out by the Digital Curator. Even so, this is probably
In my work on the Digital Curator project, I investigated these ap- the largest collection of reproductions of Central European art that can
proaches. The terms capture a collection of software tools, experiments now be handled by algorithms.
and applications that I created during my PhD at the Academy of Arts, From the summer of 2022, the Digital Curator’s collection con-
Architecture and Design in Prague. Ultimately, these tools aim to bring nov- tained 196,000 works, mostly paintings, drawings and prints from
el perspectives in respect to the study of artistic creations. Even though the holdings of 90 museums in Austria, Bavaria, the Czech Republic, and
some of these outputs are online and accessible at digitalcurator.art and Slovakia. It included reproductions of exhibits from the Albertina in Vienna,
atlas.digitalcurator.art, there are several others that do not presently the Belvedere, the Alte Pinakothek and Neue Pinakothek in Munich,
have a user interface, and thus my descriptions and examples explored the Prague City Gallery, the National Gallery in Prague, the Moravian
here will be derived from a limited sample of my work. Gallery in Brno, the Slovak National Gallery and a number of smaller
Three different experiments will be presented in this text. The first collections – from the Benedictine Abbey in Ottobeuren, West Bavaria,
one is called Curator of Art Techniques and focuses on the automated to the East Slovak Gallery in Košice.
classification of various drawings, paintings or prints. Here, an algorithm The database’s focus on Central European institutions responded
takes on the role of an expert, who classifies whether a particular canvas to the shared cultural history of the region, where artists have traveled
is an oil painting, watercolor, a woodcut, or an engraving. The second for centuries between cities, aristocratic courts, workshops, schools and
experiment titled Curator of Taste, focuses on collections that have academies, sharing their artistic style, formal elements, genres, themes
appeared under the guidance of a curator. It is a speculative attempt and motifs. It was not only the artists themselves who traveled, but also
to extrapolate the formal tastes of curators, in this case illustrated in paintings and entire collections that gradually changed owners and lo-
the painted collections of Rudolf II, and apply them to a look at later works cations. Once related works of art are now spread across hundreds of
of art produced outside the given circle. This experiment created special- museums in several independent countries, and to get a handle on even
ized neural networks that reconstructed the tastes of long-dead curators a small fraction of this heritage is beyond human capacity.
based on works that once belonged to their collections. The provocative Regarding the database’s composition, it should be highlighted
question was whether the aesthetic preferences of people from the past that despite its large size, it could not and cannot be statistically repre-
could be reconstructed in this way. sentative. There are no collections in museums that represent period art
286 287