Show simple item record

dc.contributor.authorZhu, Xiangxin
dc.contributor.authorRamanan, Deva
dc.contributor.authorFowlkes, Charless C.
dc.contributor.authorVondrick, Carl Martin
dc.date.accessioned2016-08-30T21:10:09Z
dc.date.available2016-08-30T21:10:09Z
dc.date.issued2015-03
dc.date.submitted2013-06
dc.identifier.issn0920-5691
dc.identifier.issn1573-1405
dc.identifier.urihttp://hdl.handle.net/1721.1/104080
dc.description.abstractDatasets for training object recognition systems are steadily increasing in size. This paper investigates the question of whether existing detectors will continue to improve as data grows, or saturate in performance due to limited model complexity and the Bayes risk associated with the feature spaces in which they operate. We focus on the popular paradigm of discriminatively trained templates defined on oriented gradient features. We investigate the performance of mixtures of templates as the number of mixture components and the amount of training data grows. Surprisingly, even with proper treatment of regularization and “outliers”, the performance of classic mixture models appears to saturate quickly ( ∼10 templates and ∼100 positive training examples per template). This is not a limitation of the feature space as compositional mixtures that share template parameters via parts and that can synthesize new templates not encountered during training yield significantly better performance. Based on our analysis, we conjecture that the greatest gains in detection performance will continue to derive from improved representations and learning algorithms that can make efficient use of large datasets.en_US
dc.description.sponsorshipNational Science Foundation (U.S.). (NSF IIS-0954083 and NSF DBI-1053036)en_US
dc.description.sponsorshipUnited States. Office of Naval Research. Multidisciplinary University Research Initiative (ONR-MURI N00014-10-1-0933)en_US
dc.publisherSpringer USen_US
dc.relation.isversionofhttp://dx.doi.org/10.1007/s11263-015-0812-2en_US
dc.rightsArticle is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use.en_US
dc.sourceSpringer USen_US
dc.titleDo We Need More Training Data?en_US
dc.typeArticleen_US
dc.identifier.citationZhu, Xiangxin et al. “Do We Need More Training Data?” International Journal of Computer Vision 119.1 (2016): 76–92.en_US
dc.contributor.departmentMassachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratoryen_US
dc.contributor.departmentMassachusetts Institute of Technology. Department of Electrical Engineering and Computer Scienceen_US
dc.contributor.mitauthorVondrick, Carl Martinen_US
dc.relation.journalInternational Journal of Computer Visionen_US
dc.eprint.versionAuthor's final manuscripten_US
dc.type.urihttp://purl.org/eprint/type/JournalArticleen_US
eprint.statushttp://purl.org/eprint/status/PeerRevieweden_US
dc.date.updated2016-06-30T12:07:47Z
dc.language.rfc3066en
dc.rights.holderSpringer Science+Business Media New York
dspace.orderedauthorsZhu, Xiangxin; Vondrick, Carl; Fowlkes, Charless C.; Ramanan, Devaen_US
dspace.embargo.termsNen
dc.identifier.orcidhttps://orcid.org/0000-0001-5676-2387
mit.licensePUBLISHER_POLICYen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record