Event Location: NSH 1507

Abstract: We present a discriminative shape-based algorithm for
object category localization and recognition. Our method
learns object models in a weakly-supervised fashion, without
requiring the specification of object locations nor pixel
masks in the training data. We represent object models
as cliques of fully interconnected parts, exploiting only
the pairwise geometric relationships between them. The
use of pairwise relationships enables our algorithm to successfully
overcome several problems that are common to
previously-published methods. First, our training does not
require manual labeling of object location. Thus our learning
stage is translation-invariant and robust to clutter in
the image. Second, we can evaluate features based on their
contribution as a team rather than individually. This allows
us to employ a large pool of features that, while not necessarily
discriminative in isolation, are highly discriminative
in concert. Third, the use of a fast approximate algorithm
enables us to work efficiently with pairwise relationships
among hundreds of object parts. Even though our algorithm
can easily incorporate local appearance information
from richer features, we purposefully do not use them in order
to demonstrate that simple pairwise geometric relationships
between parts can match (or exceed) the performance
of state-of-the-art object recognition algorithms.