Acts of Data
In the Acts of Data series of projects, I investigate the speculative nature of computers generating—via Generative Adversarial Networks (GANs)—new potential shapes, sounds, and spaces created through image synthesis of pre-existing publicly available datasets of 3D models. For this series, I developed several custom applications that can learn and generate new 3D images via machine learning and computer data processing.
Within the first framework ‘’Stones’’, I used the dataset provided by the “Scan the World” initiative, the largest repository that hosts scanned 3D models of historical statues from some of the world’s most prominent museums. I processed two datasets of historical artifacts: one consisting of busts and the other of full-figured statues. These datasets were then used to train custom-designed computer models, which can now independently generate entirely new 3D sculptures.
Scapes
In the framework ‘’Scapes’’, I explore the impact of environmental sounds on matter, particularly the human body and its environment, drawing from the study of soundscape ecology. The initial visual “landscapes” in this study are derived from UV (texture) maps of statues generated in the ‘’Stones’’ framework.
The generative models (statues) created in the first iteration of the project consist of two files: an OBJ file of the 3D model and a UV map, which represents the texture or landscape of the sculpture. Using machine learning techniques—specifically, interpolation in latent space—I created subtle transitions between these maps, achieving a continuous generative 3D landscape. This landscape symbolically represents the collective body of sculptures made in the previous iteration.
I chose these formations because of the historical connection between the body and landscape. In Western culture, the metaphorical use of body imagery in relation to the landscape is significant. The Renaissance metaphor that likened the earth to the human body typically viewed this relationship as one-directional—’landscape as body.’ However, in literature, the reverse—’body as landscape’—is common and persists into the machine age.
The three-video installation features each ‘’landscape’’ influenced by gathered field recordings of one of the sound categories from soundscape ecology. This field divides environmental sounds into three categories: “biophony,” “geophony,” and “anthrophony.” Biophony refers to sounds created by organisms, geophony encompasses non-biological sounds, and anthrophony includes sounds caused by humans. Through machine learning, these environmental sounds individually erode the generative landscapes, visually representing the differences in their impact on space, ecology, and the human body.
This critical approach challenges our understanding of the sensory interactions between humans and their surroundings, encouraging a deeper reflection on the often unnoticed impact of soundscapes on ourphysical and emotional well-being.
Acknowledgments:
AI and programming: Benjamin Fele | 3D animation: Luka Grčar
The project Acts of Data: Scapes was co-funded by the Ministry of Culture of the Republic of Slovenia and The Municipality Of Ljubljana