aiculedssul

Egor Kraft's Content Aware Studies (CAS) series explores the ethical issues of bias and authenticity in historical analysis using machine learning. Through AI experiments and 3D printed reproductions of ancient and prehistoric artefacts, the series examines the implications of computationally generated histories and the epistemological challenges they pose.

It is worth familiarising yourself with the project description and the photographic and video documentation that can be found on the artist's website.

While working on this project, I was responsible for creating datasets, training AI models and visualising the generated results. The datasets we created for this project consisted of two types of data. First, we collected our own custom dataset by 3D scanning the objects. Then we augmented this data with the scraped models, which are publicly available on the internet and have appropriate licences.

While working on this project I had the opportunity to work with a whole range of neural networks. While the first iterations were done with the good old DCGAN, more recently I have been working with the diffusion models and more modern GAN architectures, including those capable of working directly with 3D models by converting them to SDF.

As for the visualisation, I wrote custom Python scripts for Blender to tune and render the models generated by SDF-stylegan.