Unsupervised Active Learning For Video Annotation

Emre Demir (1) , Zehra Cataltepe (1) , Umit Ekmekci (1) , Mateusz Budnik (2) , Laurent Besacier (3, 4)

1 Istanbul University
2 LIG - Laboratoire d'Informatique de Grenoble
3 GETALP - Groupe d’Étude en Traduction Automatique/Traitement Automatisé des Langues et de la Parole
4 IUF - Institut universitaire de France

When annotating complex multimedia data like videos, a human expert usually annotates them manually. However, labeling these immense quantities of videos manually is a labor-intensive and time-consuming process. Therefore, computational methods, such as active learning are used to help annotate. In this study, we propose a cluster based unsupervised active learning approach and a new active learning method for un-supervised active learning on REPERE (Giraudel et al., 2012) video dataset, which is created for the problem of person identification in videos. Our study aims to identify who is speaking and who is on screen by using multi-modal data.

Domaines: Informatique et langage [cs.CL]

Emre Demir, Zehra Cataltepe, Umit Ekmekci, Mateusz Budnik, Laurent Besacier. Unsupervised Active Learning For Video Annotation. ICML Active Learning Workshop 2015, Jul 2015, Lille, France. ⟨hal-01350092⟩