Learning from narrated instruction videos

Learning from narrated instruction videos

ID:40636395

大小:1.43 MB

页数:13页

时间:2019-08-05

Learning from narrated instruction videos_第1页
Learning from narrated instruction videos_第2页
Learning from narrated instruction videos_第3页
Learning from narrated instruction videos_第4页
Learning from narrated instruction videos_第5页
资源描述:

《Learning from narrated instruction videos》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、LearningfromnarratedinstructionvideosJean-BaptisteAlayracyPiotrBojanowskiNishantAgrawalJosefSivicENSInriaIIITHyderabadInriaIvanLaptevSimonLacoste-JulienyInriaInriaAbstractWeaddresstheproblemofautomaticallylearningthemainstepstocompleteacertaintask,suchaschangingacartire,f

2、romasetofnarratedinstructionvideos.Thecontributionsofthispaperarethree-fold.First,wedevelopajointmodelforvideoandnaturallanguagenarrationthattakesadvantageofthecomplementarynatureofthetwosignals.Second,wecollectanannotateddatasetof57Internetin-structionvideoscontainingmorethan

3、350,000framesfortwotasks(changingcartireandCardioPulmonaryResuscitation).Third,weexperimentallydemonstratethattheproposedmodelautomaticallydiscovers,inanunsupervisedmanner,themainstepstoachieveeachtaskandlocatethemwithintheinputvideos.There-sultsfurthershowthattheproposedmodel

4、outperformssingle-modalitybaselines,demonstratingthebenefitsofjointmodelingvideoandtext.1IntroductionMillionsofpeoplewatchnarratedinstructionvideos1tolearnnewtaskssuchasassemblingIKEAfurnitureorchangingaflatcartire.Whatifmachinescouldlearnthesetasksfrommyriadsofin-structionvideo

5、sontheInternet?Suchautomaticcognitiveabilitywouldenableconstructingvirtualassistantsandsmartrobotsthatwouldhelppeopletoachievenewtasksinunfamiliarsituations.Inthiswork,weconsiderinstructionvideosanddevelopamethodthatlearnsasequenceofsteps,aswellastheirtextualandvisualrepresent

6、ations,requiredtoachieveacertaintask.Forexample,givenasetofnarratedinstructionvideosdemonstratinghowtochangeacartire,ourmethodautomaticallydiscoversconsecutivestepsforthistasksuchasloosenthenutsofthewheel,jackupthecar,removethesparetireandsoonasillustratedinFigure1.Inaddition,

7、themethodlearnsthevisualandlinguisticvariabilityofthesestepsfromnaturalvideos.arXiv:1506.09215v2[cs.CV]2Jul2015Discoveringkeystepsfrominstructionvideosisahighlychallengingtask.First,linguisticexpres-sionsforthesamestepcanhavehighvariabilityacrossvideos,forexample:“...Loosenupt

8、hewheelnutjustalittlebeforeyoustartjackingthecar...”and“...St

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。