Allowing user to interactively define and retrieve visual objects in large image or video collections is one of the main challenge of tomorrow’s multimedia search engines. In this way, we are working both on object instances retrieval, notably through large scale matching and visual query expansion, and on object categories retrieval, notably through weak geometry embedding and local relevance feedback.
Recent publications: local relevance feedback [ACM2010][LitACM2009], visual query expansion [JolyACM2009], weak geometry embedding [ICME2009] [PAMI2007]
Available Ressources: BelgaLogo dataset, for the evaluation of logo retrieval in large scale picture collections
Demos: Short video illustrating solid objects retrieval with query expansion, online VITALAS search engine including this functionality
