Birdlets: Subordinate Categorization Using Volumetric Primitives and Pose-Normalized Appearance

TitleBirdlets: Subordinate Categorization Using Volumetric Primitives and Pose-Normalized Appearance
Publication TypeConference Paper
Year of Publication2011
AuthorsFarrell, R., Oza O., Zhang N., Morariu V. I., Darrell T., & Davis L. S.
Page(s)161-168
Other Numbers3186
Abstract

Subordinate-level categorization typically rests on establishingsalient distinctions between part-level characteristicsof objects, in contrast to basic-level categorization,where the presence or absence of parts is determinative.We develop an approach for subordinate categorization invision, focusing on an avian domain due to the fine-grainedstructure of the category taxonomy for this domain. We explorea pose-normalized appearance model based on a volumetricposelet scheme. The variation in shape and appearanceproperties of these parts across a taxonomy providesthe cues needed for subordinate categorization. Trainingpose detectors requires a relatively large amount oftraining data per category when done from scratch; usinga subordinate-level approach, we exploit a pose classifiertrained at the basic-level, and extract part appearance andshape information to build subordinate-level models. Ourmodel associates the underlying image pattern parametersused for detection with corresponding volumetric part location,scale and orientation parameters. These parametersimplicitly define a mapping from the image pixels intoa pose-normalized appearance space, removing view andpose dependencies, facilitating fine-grained categorizationfrom relatively few training examples.

URLhttp://www.icsi.berkeley.edu/pubs/vision/birdletssubordinatecategorization11.pdf
Bibliographic Notes

Proceedings of the International Conference on Computer Vision (ICCV 2011), pp. 161-168, Barcelona, Spain

Abbreviated Authors

R. Farrell, O. Oza, N. Zhang, V. Morariu, T. Darrell, and L. Davis

ICSI Research Group

Vision

ICSI Publication Type

Article in conference proceedings