Computer Generated Borough Group Paintings by Ella Chedburn

CREATING DATASET:

I gathered a dataset of 600 Borough Group paintings from Borough Road Collection archives, Tate, Art UK, and Google Images. I saved each image individually which facilitates higher quality outcomes (batch downloading would mean lower quality images and less control over what gets saved – would end up being photos of the artist or other random artists work etc mixed in).

FIRST TRAINING STAGE:

The original GAN was designed to create illustrations of birds, but I fed it my Borough Group paintings instead. After 3000 steps, the images weren’t really showing any particular subject and had a high FID score (which is bad – the lower the score, the closer the generated images are to the original images. A high score is not good – mine was around 120 and showing up red/bad), so I decided to train it for 2000 more steps. The FID score decreased a little but remained high.

Since the Borough Group paintings contain differing subjects (some are landscapes, others are people, some are objects, or totally abstract) I think it is hard for the software to create figurative images. It works by looking for patterns, and the different subjects create different patterns. To avoid this, I’d have to create individual datasets for each subject (one for people, one for landscapes etc) but there’s not enough images for this (you need between 500-5000 images in each dataset).

After 3000 steps, here are 24 of the sample images:

Figures look like they’re coming through. the one below in particular reminds me of several of the figures in my dataset, particularly ‘Standing Figure’ by Dorothy Mead. So I trained for longer to see if these figures would emerge fully.

Left: computer-generated, Right: Dorothy Mead

SECOND TRAINING STAGE:

I did 2000 more steps, so at 5000 steps the sample images looked like this:

At this point I generated 50 random full quality images. Impression of figures was no longer there in the ones I expected, although new figures did appear. I felt that there was no need to train further, as the figures seemed to get lost the more they’re trained.

EXPORTING IMAGES:

The software generates thousands and thousands of new images, so finding/saving the best ones can become a mammoth task! I can choose the level of truncation -- closer to 0 = less diverse but very realistic and 1 = more diverse but maybe deviating from the rest of the sample. I decided to go for truncation of around 0.8 to get a diverse range of pictures. Then I could scroll through the pictures in different directions to select individual favourites, or get the software to generate a random sample to remain objective.

CREATING A VIDEO:

I selected different images to merge together to create an interpolation video. I included abstract works, landscapes, and figures.

Creating its own signature style of quadrangular structures:

Seemed to keep getting preoccupied with painting these grey blobs!

COMPARISONS:

Left: computer generated. ‘Oxen Ploughing – Spain’ by Dennis Creffield

Left: computer generated. Right: ‘Bent Figure’ by Edna Mann

Left: computer generated. Right: ‘Tregor and Tregoff, Cornwall’ by David Bomberg

Left: computer generated. ‘Oxen Ploughing – Spain’ by Dennis Creffield

Left: computer generated. Right: ‘Children's Swimming Pool, Friday Evening’ by Leon Kossoff

Left: computer generated. Right: ‘From Willesden Green, Autumn’ by Leon Kossoff

Left: computer generated. Right: Standing Nude by Dennis Creffield

Left: computer generated. Right: ‘Seated Nude’ by Dennis Creffield

PROCESSING AND INPUT:

These are computer generated Borough Group paintings created by artist Ella Chedburn (@ellachedburn.art). Bomberg famously said he hoped his students would find the "spirit in the mass". By using machine learning software to generate paintings from a dataset of over 600 Borough Group images, perhaps the “spirit in the mass” has been found?

The generated artworks are primarily abstract with some landscapes and impressions of figures coming through. The software’s loosely painted nudes and linear abstract compositions could be lost Dennis Creffield paintings, while certain landscapes are eerily close to Bomberg’s hills and mountains. Although many images have a convincing texture of thick layered paint, several expose their computerised nature with their smoothness. However, the overall aesthetic could successfully pass as a member of the Borough Group: loose angular brushstrokes and mostly desaturated tones with occasional shots of bold warm pigment. Just like a true artist, the software has even begun to create its own signature style of large cross-hatched structures and mysterious dark grey blobs.

Do you think A.I. could be the Borough Group’s newest member?

ARTIST BIO

Ella Chedburn is a Fine Art (BA Hons) graduate from Kingston School of Art. Her practice squeezes nature through a digital lens: scanning trees, coding virtual reality forests, wiring plants to Arduinos, and projecting into woodland. She playfully misuses technology, encouraging glitches to run throughout her digital work and mimicking them in her prints and books. Over the past year, misinformation and context-twisting has been a particular driving force; sparking the creation of her dreamy algorithmic narratives, mystical inventions, and fictional documentaries blending science with folklore. Most recently, she has been creating computer-generated paintings for the Borough Road Collection and William Morris Society, questioning whether ‘the artist’ can be found within the machine.

The Spirit in the Mass in Dorothy Mead’s Paintings

by Alice Mcleod-Bishop

Dorothy Mead (1928-1975) joined the Borough Group in 1946 as an original member of the group which was dedicated to portraying David Bomberg’s (1890-1957) Modernist teaching methods and his philosophy known as the Spirit in the Mass, until the group’s dissolution in 1951. The notion of Spirit in the Mass was primarily about the connection between art and wider life, where he aimed to capture someone as they are in the world rather than as a subject. He understood the ‘self’ as a conditional relationship in consideration of its surroundings, taking into account the artist’s perception of the subject, as well as the phenomenology (the experience of experiencing something) of the subject – to capture what it is like for the sitter, landscape, cityscape etc., to be and exist as itself rather than solely the artist’s depiction of what it might look like at face-value. The idea of mass specifically relates to the synthesis of thought and feeling: the artist must assume ignorance about the subject whilst not being ignorant, and show the world as we see it, as uncertainty. Bomberg believed that modernist drawing is seen “as a deliberate distortion of optical truth”[i], and thus one must be classically trained if aiming to distort; yet if the artist is concerned with how things ‘feel’, then what they draw will naturally be a subjective and therefore distorted view of reality. 

This is what drove Bomberg to teach his anti-establishment methods, which in turn meant he was met with distaste and disapproval from the contemporary art community and was not recognised as a legitimate teacher. His views were widely misunderstood due to the lack of clarity surrounding his ideas and the somewhat confusing way in which they are worded in the Bomberg Papers. Contemporary critics and teachers failed to ascertain what Bomberg’s philosophy entailed, and to this day his teachings are confused by many[ii]. It is highly difficult to truly grasp exactly what Bomberg aimed to portray in his classes and what is the Spirit in the Mass; it seems that only those who were his students managed to understand Bomberg’s revolutionary methods and accurately capture what Bomberg perceived to be the Spirit in the Mass. 

One of these students was Dorothy Mead, who was especially dedicated to Bomberg’s methods and was forced to leave the Slade art school because of her allegiance to the British Modernist artist. Bomberg’s influence on Mead is evident in her paintings, particularly her depictions of bodies, figures and landscapes: stylistically, Mead’s work is reminiscent of Bomberg’s post-war paintings, using thick expressive brush strokes and dark colours to portray a sense of the subject and its essence, an idea that was integral to Bomberg’s philosophy. It is difficult to delineate precisely how far Mead was able to capture the Spirit of her subjects, since we cannot know in detail her relationship with what and who she painted, nor how far she allowed her subjectivity to distort her reality. One can assume with some confidence however that she aimed to exact the Spirit of her subjects through her use of colour and the presentation of form in her works. In her 1955 work Portrait[iii]  (left), she uses bold yellow and blue brush strokes over a dark red and brown background to create the suggestion of an anonymous sitter. Compared to her undated self-portrait titled Self-Portrait[iv](right), it is clear she perceived herself quite differently to her sitter in Portrait. The use of duller colours and more succinct brush strokes in Self-Portrait might suggest she had a less-than passionate view of herself, while the vibrant, even violent feel to Portrait implies a potentially tumultuous or impassioned view of her unnamed sitter. This short comparison highlights how Mead attempted to portray her perception of her subjects and thus how she interpreted Bomberg’s philosophy of the Spirit in the Mass; the different stylistic techniques in her depiction of herself compared to that of her subject is evidence of an attempt to capture the essence and phenomenology of the individual.

[i][i] Roy Oxlade, Bomberg Papers: The Spirit in the Mass, a commentary, together with transcriptions of various previously unpublished notes, p. XIV (introduction), Royal College of Art, 1980

[ii] Cliff Holden, Bomberg’s Teaching – Some Misconceptions, p.3, 2004, Cliff Holden

[iii] Dorothy Mead, Portrait, 1955, Borough Road Gallery Archive

[iv] Dorothy Mead, Self-Portrait, (undated), Borough Road Gallery Archive

An archive of direction

This is an archive of direction. Or more specifically an archive of the directional marks in the Borough Road collection. 

Anybody who has painted knows of the breathless moment when the brush meets the surface.

Cut out, assembled and arranged according to direction, the process of categorising and re-archiving the lines gives a new room for these painterly moments to exist. 

Collectivised and cut from any pictorial context but their own presence, the lines still seems to me to be powerful memories of the artists at work; their commitment, decisions and physical presence. 

Archive of Direction by Johanna Bolton

Detail of an Archive of Direction by Johanna Bolton

Johanna Bolton: Research Image of an Archive

Over the weeks, I have been looking at the works in the collection in their context of an archive. I have found these communal themes or qualities that could be used as categories to arrange them by:

Lines - direction
Marks: lines - dashes - dots
Marks: patches - blotches -dots
Structure - angles - rounds
Surface: texture - thickness - energy or crunchiness
Grids
Colour: greys, yellow-pink, red-orange, blues
Motif: bodies, city-scapes, portraits, vases etc...
Narrative: teaching, trauma, time, instruction, intention, choice
Reaction: breathing, movement

From the research: A list of motives, and how often they occur