Trained on specific keywords, subjects and descriptions of its partners EPA and EFE, Videre AI aims to revolutionise video analysis practises. Archiving news outlets might never be the same thanks to the remarkable results achieved in terms of tagging and captioning the partners’ online content.
Overall: They trained their Multi-Modal Transformer tagging model with specific keywords from their partners EFE/epa and created a corresponding benchmark model. Then, they compared those results with out-of-the-box predictions. They present excellent results as their AI managed to recognise: keywords, IPTC subject types/matters which allows to perform a hierarchical classification of the videos, and abstract captioning of the same.