I will talk about models for fine-grained categorization, focusing on determining botanical species from scanned leaf images. The strategy is to focus attention around several vantage points, which is the approach taken by botanists. We introduce "vantage feature frames" consisting of two components: a set of coordinate systems centered at the most discriminating local viewpoints for the generic object class and a set of category-dependent features computed in these frames. Based on an underlying coarse-to-fine hierarchy, categorization then proceeds from coarse-grained to fine-grained using node classifiers associated with the frames. Finally, I will also discuss a new performance criterion motivated by applications: report a subset of species whose expected size is minimized subject to containing the true species with high probability. Everything will be illustrated on multiple leaf datasets with comparisons to existing methods.
Copyright. All Rights Reserved.