This paper examines how far we are towards automatic
extraction of building features, by comparing two pipelines from
image data to building features: an ideal pipeline, based on the
requirements of an on-going project, and a realistic pipeline, based on
current computer vision technologies.