Paper Title:Recognition, Description and Localisation of Images Using WS-MRF-SIBP


This paper proposed a WS-MRF-SIBP model to learn the weakly labeled images. The object, attribute and background appearances, object – attribute association and their locations from realistic weakly labeled images including multiple objects with cluttered background are learned from the images. Then a novel weakly supervised Bayesian model is formulated to learn and exploit spatial coherence and factor co-occurrence. Once learned from weakly labeled data, this model performs various tasks including semantic segmentation, image description and image query.

Keywords:Weakly supervised learning, object-attribute association, semantic segmentation, non-parametric Bayesian model, Indian Buffet Process