cvpr18-Unifying Identification and Context Learning for Person Recognition

cvpr18-Unifying Identification and Context Learning for Person Recognition

ID:40352192

大小:2.43 MB

页数:9页

时间:2019-07-31

cvpr18-Unifying Identification and Context Learning for Person Recognition_第1页
cvpr18-Unifying Identification and Context Learning for Person Recognition_第2页
cvpr18-Unifying Identification and Context Learning for Person Recognition_第3页
cvpr18-Unifying Identification and Context Learning for Person Recognition_第4页
cvpr18-Unifying Identification and Context Learning for Person Recognition_第5页
资源描述:

《cvpr18-Unifying Identification and Context Learning for Person Recognition》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、UnifyingIdentificationandContextLearningforPersonRecognitionQingqiuHuang,YuXiong,DahuaLinCUHK-SenseTimeJointLab,TheChineseUniversityofHongKong{hq016,xy017,dhlin}@ie.cuhk.edu.hkVisualContextSocialContextAbstractEvent-PersonHeadDespitethegreatsuccessoffacer

2、ecognitiontechniques,FacerecognizingpersonsunderunconstrainedsettingsremainsFaceGBodyPerson-Personchallenging.Issueslikeprofileviews,unfavorablelighting,andocclusionscancausesubstantialdifficulties.PreviousUpperBodyInferenceworkshaveattemptedtotacklethispr

3、oblembyexploitingInferenceInferenceLearningthecontext,e.g.clothesandsocialrelations.Whileshowing???KateLeonardoKatepromisingimprovement,theyareusuallylimitedintwoim-portantaspects,relyingonsimpleheuristicstocombinedif-ferentcuesandseparatingtheconstructi

4、onofcontextfrompeopleidentities.Inthiswork,weaimtomovebeyondsuchlimitationsandproposeanewframeworktoleveragecon-textforpersonrecognition.Inparticular,weproposeaRe-Figure1:Personrecognitionunderunconstrainedsettingsre-gionAttentionNetwork,whichislearnedto

5、adaptivelycom-mainsaverychallengingproblem.Inferencepurelybyfacerecog-binevisualcueswithinstance-dependentweights.Wealsonitiontechniqueswouldfailinmanycases.Weproposeaframe-developaunifiedformulation,wherethesocialcontextsareworktotacklethisproblem,whichc

6、ombinesvisualcontextwithlearnedalongwiththereasoningofpeopleidentities.Theseadaptiveweightsandunifiespersonrecognitionwithsocialcon-modelssubstantiallyimprovetherobustnesswhenworkingtextlearning.withthecomplexcontextualrelationsinunconstraineden-vironment

7、s.Ontwolargedatasets,PIPA[27]andCastInThedifficultiesaboveareessentiallyduetothefactthatMovies(CIM),anewdatasetproposedinthiswork,ourfacialappearanceishighlysensitivetoenvironmentalcon-methodconsistentlyachievesstate-of-the-artperformanceditions.Totacklet

8、hisproblem,anaturalideaistoleverageundermultipleevaluationpolicies.anotherimportantsourceofinformation,namelythecon-text.Itisourcommonexperiencethatwecaneasilyrec-ognizeafamiliarpersonbylookingatthewearing,thesur-1.Introdu

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。