AI search
A collection of over 30 hours of films on the environment, fully accessible via scene, text and audio search.
This is a pre-trained model that we chose to test on a corpus of more than 200 films representing 33 hours of moving images of all types: 8mm, 9.5mm, and 16mm films; VUS, UMATIC, Beta. What is interesting is that this corpus, which is specific to the environment and for which human documentation emphasises aspects related to this theme, allows AI to obtain results from very simple searches, such as a smile, a bird, a table... Objects and shapes that documentalists rarely or never focus on, thus providing added value to traditional documentation.
As part of the MeFiTer project carried out between 2022 and 2025 by the Diazinteregio Association and its members, the focus has been on using artificial intelligence to enhance collections of amateur, institutional and activist films. While the initial experiments conducted by the Cinémathèque de Bretagne in partnership with the IMT in Brest focused on assisted documentation and predictive image dating, these studies revealed that significant machine learning work would be required to obtain meaningful results in terms of documentation. Object recognition works, but the contextualisation of image collections that are small in size compared to the mass of contemporary media is more uncertain.
We chose to use AI for what it already masters: scene recognition and OCR. With the support of Archipanion, which had already worked on Swiss archive collections, we tested a pre-trained model on our moving images. You can see the results of this work at archipanion.amorce.eu
We hope you enjoy discovering it and please feel free to send us your feedback at contact {at} diazinteregio.org
