Sparselet Models for Efficient Multiclass Object Detection

TitleSparselet Models for Efficient Multiclass Object Detection
Publication TypeConference Paper
Year of Publication2012
AuthorsSong, H. Oh, Zickler S., Althoff T., Girshick R., Geyer C., Fritz M., Felzenszwalb P., & Darrell T.
Page(s)802-815
Other Numbers3360
KeywordsDeformable Part Models, Object Detection, Sparse Coding
Abstract

We develop an intermediate representation for deformablepart models and show that this representation has favorable performance characteristics formulti-class problems when the number of classes is high. Our model uses sparse coding of partfilters to represent each filter as a sparse linear combination of shared dictionary elements. Thisleads to a universal set of parts that are shared among all object classes. Re- construction of theoriginal part filter responses via sparse matrix-vector product reduces computation relative toconventional part filter convolutions. Our model is well suited to a parallel implementation, andwe report a new GPU DPM implementation that takes advantage of sparse coding of part filters. Thespeed-up offered by our intermediate representation and parallel computation enable real-time DPMdetection of 20 different object classes on a laptop computer.

Acknowledgment

S. Zickler and C. Geyer were supported by DARPA con-tract W911NF-10-C-0081. P. Felzenszwalb and R. Girshick were supported inpart by NSF grant IIS-0746569. T. Darrell was supported by DARPA contractW911NF-10-2-0059, by NSF awards IIS-0905647, IIS-0819984, and support fromToyota and Google.

URLhttps://www.icsi.berkeley.edu/pubs/vision/sparseletmodels12.pdf
Bibliographic Notes

Proceedings of the 12th European Conference on Computer Vision (ECCV 2012), pp. 802-815, Firenze, Italy

Abbreviated Authors

H. O. Song, S. Zickler, T. Althoff, R. Girshick, C. Geyer, M. Fritz, P. Felzenszwalb, and T. Darrell

ICSI Research Group

Vision

ICSI Publication Type

Article in conference proceedings