Open access
Author
Date
2023-12-01Type
- Other Conference Item
ETH Bibliography
yes
Altmetrics
Abstract
Artificial intelligence (AI), computer vision, object and image recognition or artificial neural networks are currently being used excessively in the media. AI applications are increasingly penetrating the consumer sector, often unnoticed. Examples include facial recognition in mobile phone cameras, pattern recognition in diagnostic medicine, image processing on social media and visual recognition in autonomous driving. This is also accompanied by a great deal of uncertainty and ignorance.
"Autotagging" is the automatic assignment of keywords/tags to images based on the image content without human intervention. The digital asset management system Canto Cumulus, which has been in use in the ETH Library's image archive for over 17 years, has offered the computer vision software "Clarifai", trained using artificial neural networks, as a plugin for autotagging since 2021.
The image archive has had its entire image stock of over 1 million images autotagged with Clarifai's General base model to supplement the hierarchical keyword catalogue. The time required was considerable. The approximately 4600 autotags were then analysed to assess the quality of the AI recognition. This means that a large historical database, which was not only enriched with AI but also validated, is now available for further applications in the digital humanities and for generating new AI models for historical image material. Experiences, advantages and disadvantages and, in particular, the qualitative evaluation of the autotags will be presented at EVA Berlin. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000645039Publication status
publishedPublisher
ETH-BibliothekEvent
Subject
AI; SACHERSCHLIESSUNG + SACHKATALOGISIERUNG (BIBLIOTHEKSWESEN); künstliche Intelligenz; BILDARCHIVE, BILDSAMMLUNGENOrganisational unit
00060 - Abt. ETH-Bibliothek / ETH-Bibliothek
Related publications and datasets
Documents: http://hdl.handle.net/20.500.11850/645197
Notes
Conference lecture held on December 1, 2023.More
Show all metadata
ETH Bibliography
yes
Altmetrics