We address the problem of localisation of objects as bounding boxes in images
and videos with weak labels. This weakly supervised object localisation problem
has been tackled in the past using discriminative models where each object
class is localised independently from other classes.