ABSTRACT
One of the central questions in object categorisation is how the shape variability of different category members can be accounted for. It was proposed recently that the shape variability of objects from a basic level category usually can be described well with continuous deforming shape transformations. A database of basic level categories was created by constructing morphable 3D models of exemplars from 29 different categories, covering both biological and artefact categories. New members of each category can be produced by morphing between two category exemplars. Categorisation performance deteriorated systematically with increased shape transformation. The systematic dependency of categorisation performance on the amount of shape transformation is a very robust finding. The findings strongly indicate image-based instead of abstract or propositional category representations, but are compatible with a role of object parts in categorisation.
