Unland Imaginaries

Stereoscopes that became popular in the 19th Century use two photographs taken an eye-width apart to create the illusion of 3D viewing. The stereo images in Unland Imaginaries are created from single (monocular) photographs originally captured by Google street view from either side of the buffer zone in Nicosia. Machine learning techniques were used to estimate depth and create each stereoscopic pair.
The stereo pair on the right hand side of each scene were altered using text-to-image machine learning drawn from interviews of people displaced from Nicosia in 1974. These were used to form instructions to ‘paint in’ elements to the scenes.

Stereoscopic viewing devices are available in the gallery for visitors to view the original stereoscope pair, the pair of images generated using text-to-image machine learning AI, or a combination of the two.

Below: click on individual images to enlarge…

Nicosia buffer zone (North)

Depth matte for stereo pair pictured above of Nicosia buffer zone (North)

Above: 2 sets of stereo pairs North of Unland Imaginaries and their corresponding depth mattes. The first two images are based on an initial monocular Google Streetview image. This is then turned into a stereo image using a depth recognition programme (the depth matte of which corresponds to the greyscale image on the left immediately below it). The second stereo pair use machine learning text to image inpainting to alter the initial image. With the above set, the instructions for inpainting (the prompts) use natural language derived from interviews with people talking about when they lived in the buffer zone prior to 1974.

Nicosia buffer zone (South)

Depth matte for stereo pair pictured above of Nicosia buffer zone (North)

Distance: Depth matte street views real or imagined – reveal the background digital processes used to make the stereoscopic series Unland Imaginaries. They are part of the process of estimating depth using machine learning techniques, together with inpainting techniques – the painting-in of details into part of an image. The original images these are based on, were captured and processed by google street view cameras. The technique on display here is also part of scene-understanding that has been used in autonomous robots such as self-driving cars.

A selection of stereo pairs:

Nicosia buffer zone (South)

I inputted the the left stereo pair in this scene into a second machine learning programme – Amazon’s ‘Image Rekognition’ programme. This analysed the buffer zone image on the left and came up with a list of 100 labels for the image, based on confidence. Susprisingly, what scored quite highly for this buffer zone architecture was PATIO, SWIMMING POOL, GARDEN, VERANDA. The ouptut from Amazon’s ‘Image Rekognition’ analysis was then fed into the text to image inpainting programme which produced the stereo pair on the right.

Nicosia buffer zone (North)

Nicosia buffer zone (South)

The locations of the two scenes above (the top images facing from North to South and the bottom images facing from South to North) were partly chosen for being as close as possible to each other from the limited journeys collected by Google Streetview cars (in 2022 at least).