WG3 – Automatic Generation fo Image and Video Descriptions

WG3 focuses on methods for annotating, labeling and describing visual data, including integration of language technologies in annotating visual data using suitable weakly supervised machine learning models, inference models that take into account language and visual constraints, latent class models for coping with variant low level features, alignment models, models that detect complementarity of the vision and text content, and text generation methods.

WG3 Leader: Luc van Gool; Deputy: Erkut Erdem.