AI for Image Collections: A Hands-On Workshop
Welcome to a hands-on introduction to AI methods for exploring digital image collections, presented by KBLab within the national infrastructure Huminfra.
A visual theme cluster generated with CLIPtopic.
Why this workshop
Many cultural-heritage organisations and researchers hold large image collections with limited or uneven metadata. That makes search and overview difficult – often a “needle in a haystack” problem. Without descriptions to make them searchable, these images remain difficult to access. Recent AI tools can be used to address this challenge. In particular, the multimodal model CLIP enables novel text-and-image search – even when metadata is sparse. This workshop shows how CLIP can be combined with topic modelling to surface visual themes at scale.
CLIPtopic integrates CLIP’s multimodal features with the clustering power of topic modelling. Discover how multimodal topic modelling can open new, more accessible ways of exploring image collections.
What you’ll learn
After a short intro to CLIP and topic models, we’ll walk through the steps to build a CLIPtopic model and then discuss how to interpret the results. You’ll gain enough practical understanding – and curiosity – to test this approach in your own projects or at your memory institution. Participants can reuse the provided script with their own data after the workshop.
Practicalities
When
Tuesday 21 October, 10:00–12:00, online (Zoom).
Who is it for?
Primarily researchers and heritage professionals, but all are welcome – no programming skills required. The workshop language (Swedish or English) will be decided based on participants; mixing is possible.
Requirements
Free of charge with support from Huminfra. To follow the hands-on part you’ll need a Google account to log into Google Colab (our interactive environment).
How to apply
Email: kblabb@kb.se by Friday 17 October.
Places are limited. While open to all (including researchers and Master’s students), priority will be given to PhD candidates and heritage professionals.
About KBLab
KBLab is a national infrastructure for digital research at the National Library of Sweden (KB). Beyond supporting large-scale analysis of KB’s collections by digital research projects in the humanities and social science, we use the library’s vast data resources to train and release open-source AI models that are being used by a wide range of actors in the public sector and beyond. You can read more about our development projects within AI and data science on our blog.
More information
About Huminfra
Huminfra is a Swedish national infrastructure supporting digital and experimental research in the Humanities by providing users with a single entry point for finding existing Swedish materials and research tools, as well as developing national method courses. You can access these resources via Huminfra’s web portal here.