PDF version of this document

METHOD

We automatically build an appearance model from the correspondences identified by a registration. We then synthesise images from the model and perform a cross-imageset comparison to assess model/registration quality.