AI-Driven Classification of a Design Photographic Archive

The paper presents a workflow for deploying an Artificial Intelligence (AI) classification of a previously unclassified photographic collection, the Design Archive's glass plate negatives. This involved fine-tuning the DinoV2 self-supervised image retrieval system with a domain-expert taxonomy to classify approximately 10K images within 40 classes. As such, it addresses challenges relevant to the curation, analysis and discovery of large-scale visual collections. A 3D visualisation was implemented for users to access the outputs presenting images as data points using the Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) to project the embeddings of the neural network. The paper demonstrates the advantages of this approach and reflects how users can participate in the AI processes making them more transparent and trustable.