According to findings by scientists at MIT it takes less than a few milliseconds for our brains to process an image we see. The ability to connect with an artwork involves attention to detail, our aesthetic sensibilities and is deeply personal.

However, in this age of sensory and visual overload we have to ask whether we’re ever in front of an image for long enough? Long enough, that is, to register the synthesis that occurs between the viewed and the viewer and results in the formation of what we call perception. Are we too easily distracted? Is all we get now in terms of meaning and substance just rudimentary content recognition, engaging for such short periods of time – mere momentary glimpses?

I believe that the growing habit of skimming directly relates to our struggle to hold certain encounters in our long-term memories. This inability to hold on to facts and experiences, I want to suggest, limits the extent of our imagination and can be a challenge to our emotional wellbeing.

In an era of visual abundance, this work aims to de-construct the act of photographic representation, encouraging us to rediscover the processes and understand what we “see”.

Conceived as a multi-part project, the first part consists of a series of plates, reminiscent of Ophthalmologists’ charts. They describe photographic images and encounters, rather than depict them. The second part of the project then sees the feeding of the textual plates into a GAN based machine-learning system. The ‘Story-to-Image’ Generator is an AI system that is able to visualise what is described in a text. The neural networks then conceive unique pictures based on the image descriptions.

We based the project on the AttnGAN approach, implemented in PyTorch machine learning framework. In particular, AttnGAN uses an attentional generative network that synthesises progressively refined details in different regions of the image. It does so by observing the structure of the prose and the relevant words in the text description.

–

–

AI Modelling:
The Text-to-Image Generator is an AI system that is able to visualise what is described in a text.

–

Plates:
Giclée prints on Hahnemühle 308 gsm photorag, 78x78cm.