The starting point was Joseph Kosuth’s work, «One and Three Chairs,» where a single object exists simultaneously in three forms: as a real thing, a photograph, and a textual description. I was interested in applying this principle to a living organism and adding another layer—its digital representation through data. I chose the fly as the object; it is easily recognizable by a small set of characteristics while also being well-suited for strong visual transformation.
The connectome of the fruit fly—a map of the neural connections in its brain—is publicly available. The project utilized data from FlyWire v783, which contains the actual identifiers for the neurons and the connections between them.This led to an alternative way of describing an object: not through a photograph or text, but through the structure of its nervous system.I wanted to investigate whether this structure could be used not merely as visual material, but as part of a computational system.The project idea evolved to this:fly images → real neural connections → training → averaged representation → new visualization
The program is written in Python and utilizes FlyWire data from the file proofread_connections_783.feather. A subset of real neurons and the connections between them is selected from the full connectome to construct a working network. Images of flies are loaded into this network: the program extracts visual features, converts them into a numerical signal, and passes the signal through the network.
During training, the system compares the predicted line direction with the correct one and adjusts the trainable output layer. A set of various photographs allows the formation of an averaged image, which is then rendered with lines and saved in PNG, SVG, and GIF formats. The original FlyWire connections remain unchanged—they serve as the foundation for the calculations, and the training occurs on top of them.Following this, the resulting abstract image is uploaded to the Recraft neural network, which uses it to generate the final image.
The project presents a single object through multiple modes of representation: photograph, text, neural data, and machine interpretation. The central element of the work is a proprietary program that uses real FlyWire data as the structure for a computational network trained to work with visual imagery.
