Bastian Leibe Jiri Matas Nicu Sebe Max Welling (Eds.) 0 Computer Vision – 1 9 9 S ECCV 2016 C N L 14th European Conference Amsterdam, The Netherlands, October 11–14, 2016 Proceedings, Part VI 123 Lecture Notes in Computer Science 9910 Commenced Publication in 1973 Founding and Former Series Editors: Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen Editorial Board David Hutchison Lancaster University, Lancaster, UK Takeo Kanade Carnegie Mellon University, Pittsburgh, PA, USA Josef Kittler University of Surrey, Guildford, UK Jon M. Kleinberg Cornell University, Ithaca, NY, USA Friedemann Mattern ETH Zurich, Zurich, Switzerland John C. Mitchell Stanford University, Stanford, CA, USA Moni Naor Weizmann Institute of Science, Rehovot, Israel C. Pandu Rangan Indian Institute of Technology, Madras, India Bernhard Steffen TU Dortmund University, Dortmund, Germany Demetri Terzopoulos University of California, Los Angeles, CA, USA Doug Tygar University of California, Berkeley, CA, USA Gerhard Weikum Max Planck Institute for Informatics, Saarbrücken, Germany More information about this series at http://www.springer.com/series/7412 Bastian Leibe Jiri Matas (cid:129) Nicu Sebe Max Welling (Eds.) (cid:129) – Computer Vision ECCV 2016 14th European Conference – Amsterdam, The Netherlands, October 11 14, 2016 Proceedings, Part VI 123 Editors Bastian Leibe NicuSebe RWTH Aachen University of Trento Aachen Povo- Trento Germany Italy JiriMatas MaxWelling Czech TechnicalUniversity University of Amsterdam Prague 2 Amsterdam Czech Republic TheNetherlands ISSN 0302-9743 ISSN 1611-3349 (electronic) Lecture Notesin Computer Science ISBN 978-3-319-46465-7 ISBN978-3-319-46466-4 (eBook) DOI 10.1007/978-3-319-46466-4 LibraryofCongressControlNumber:2016951693 LNCSSublibrary:SL6–ImageProcessing,ComputerVision,PatternRecognition,andGraphics ©SpringerInternationalPublishingAG2016 Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartofthe material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodologynow knownorhereafterdeveloped. Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse. Thepublisher,theauthorsandtheeditorsaresafetoassumethattheadviceandinformationinthisbookare believedtobetrueandaccurateatthedateofpublication.Neitherthepublishernortheauthorsortheeditors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissionsthatmayhavebeenmade. Printedonacid-freepaper ThisSpringerimprintispublishedbySpringerNature TheregisteredcompanyisSpringerInternationalPublishingAG Theregisteredcompanyaddressis:Gewerbestrasse11,6330Cham,Switzerland Foreword Welcome to the proceedings of the 2016 edition of the European Conference on ComputerVisionheldinAmsterdam!ItissafetosaythattheEuropeanConferenceon ComputerVisionisoneofthetopconferencesincomputervision.Itisgoodtoreiterate the history of the conference to see the broad base the conference has built in its 13 editions. First held in 1990 in Antibes (France), it was followed by subsequent con- ferences in Santa Margherita Ligure (Italy) in 1992, Stockholm (Sweden) in 1994, Cambridge (UK) in 1996, Freiburg (Germany) in 1998, Dublin (Ireland) in 2000, Copenhagen (Denmark) in 2002, Prague (Czech Republic) in 2004, Graz (Austria) in 2006,Marseille(France)in2008,Heraklion(Greece)in2010,Florence(Italy)in2012, and Zürich (Switzerland) in 2014. For the 14th edition, many people worked hard to provide attendees with a most warm welcome while enjoying the best science. The Program Committee, Bastian Leibe, Jiri Matas, Nicu Sebe, and Max Welling, did an excellent job. Apart from the scientificprogram,theworkshopswereselectedandhandledbyHervéJégouandGang Hua, andthetutorials byJacob VerbeekandRita Cucchiara.Thanksfor thegreatjob. The coordination with the subsequent ACM Multimedia offered an opportunity to expand the tutorials with an additional invited session, offered by the University of Amsterdam and organized together with the help of ACM Multimedia. Ofthemanypeoplewhoworkedhardaslocalorganizers,wewouldliketosingleout MartinedeWitoftheUvAConferenceOffice,whodelicatelyandefficientlyorganized themainbody.AlsothelocalorganizersHamdiDibeklioglu,EfstratiosGavves,Janvan Gemert,ThomasMensink,andMihirJainhadtheirhandsfull.Asavenue,wechosethe RoyalTheatreCarrélocatedonthecanalsoftheAmstelRiverindowntownAmsterdam. SpaceinAmsterdamissparse,soitwasalittletighterthanusual.Theuniversitylentus their downtown campuses for the tutorials and the workshops. A relatively new thing wastheindustryandthesponsorsforwhichRonaldPoppeandPeterdeWithdidagreat job, while Andy Bagdanov and John Schavemaker arranged the demos. Michael WilkinsontookcaretomakeYomKippurascomfortableaspossibleforthoseforwhom itisanimportantday.WethankMarcPollefeys,AlbertodelBimbo,andVirginieMes fortheiradviceandhelpbehindthescenes.Wethankalltheanonymousvolunteersfor theirhardandprecisework.Wealsothankourgeneroussponsors.Theirsupportisan essentialpartoftheprogram.Itisgoodtoseesuchalevelofindustrialinterestinwhat our communityis doing! Amsterdam does not need any introduction. Please emerge yourself but do not drown in it, have a nice time. October 2016 Theo Gevers Arnold Smeulders Preface Welcome to the proceedings of the 2016 European Conference on Computer Vision (ECCV 2016) held in Amsterdam, The Netherlands. We are delighted to present this volume reflecting a strong and exciting program, the result of an extensive review process.Intotal,wereceived1,561papersubmissions.Ofthese,81violatedtheECCV submission guidelines or did not pass the plagiarism test and were rejected without review. We employed the iThenticate software (www.ithenticate.com) for plagiarism detection.Oftheremainingpapers,415wereaccepted(26.6%):342asposters(22.6%), 45asspotlights(2.9%),and28asoralpresentations(1.8%).Thespotlights–short,five- minute podium presentations – are novel to ECCV and were introduced after their successattheCVPR2016conference.Alloralsandspotlightsarepresentedaspostersas well. The selection process was a combined effort of four program co-chairs (PCs), 74areachairs(ACs),1,086ProgramCommitteemembers,and77additionalreviewers. As PCs, we were primarily responsible for the design and execution of the review process. Beyond administrative rejections, we were involved in acceptance decisions onlyintheveryfewcaseswheretheACswerenotabletoagreeonadecision.PCs,as iscustomaryinthefield,werenotallowedtoco-authorasubmission.Generalco-chairs andotherco-organizersplayednoroleinthereviewprocess,werepermittedtosubmit papers, and were treated as any other author. Acceptance decisions were made by two independent ACs. There were 74 ACs, selectedbythePCsaccordingtotheirtechnicalexpertise,experience,andgeographical diversity(41fromEuropean,fivefromAsian,twofromAustralian,and26fromNorth Americaninstitutions).TheACswereaidedby1,086ProgramCommitteemembersto whom papers were assigned for reviewing. There were 77 additional reviewers, each supervised by a Program Committee member. The Program Committee was selected fromcommitteesofpreviousECCV,ICCV,andCVPRconferencesandwasextended onthebasisofsuggestionsfromtheACsandthePCs.HavingalargepoolofProgram Committee members for reviewing allowed us to match expertise while bounding reviewer loads. Typically five papers, but never more than eight, were assigned to a Program Committee member. Graduate students had a maximum of four papers to review. TheECCV2016reviewprocesswasinprincipledouble-blind.Authorsdidnotknow revieweridentities,northeACshandlingtheirpaper(s).However,anonymitybecomes difficult to maintain as more and more submissions appear concurrently on arXiv.org. This was not against the ECCV 2016 double submission rules, which followed the practice of other major computer vision conferences in the recent past. The existence of arXiv publications, mostly not peer-reviewed, raises difficult problems with the assessment of unpublished, concurrent, and prior art, content overlap, plagiarism, and self-plagiarism.Moreover,itunderminestheanonymityofsubmissions.Wefoundthat notallcasescanbecoveredbyasimplesetofrules.Almostallcontroversiesduringthe review process were related to the arXiv issue. Most of the reviewer inquiries were VIII Preface resolvedbygivingthebenefitofthedoubttoECCVauthors.However,theproblemwill havetobediscussedbythecommunitysothatconsensusisfoundonhowtohandlethe issues brought bypublishing on arXiv. Particular attention was paid to handling conflicts of interest. Conflicts of interest between ACs, Program Committee members, and papers were identified based on the authorship of ECCV 2016 submissions, on the home institutions, and on previous col- laborationsofallresearchersinvolved.Tofindinstitutionalconflicts,allauthors,Program Committee members, and ACs were asked to list the Internet domains of their current institutions. To find collaborators, the Researcher.cc database (http://researcher.cc/), fundedbytheComputerVisionFoundation,wasusedtofindanyco-authoredpapersin theperiod2012–2016.Wepre-assignedapproximately100paperstoeachAC,basedon affinity scores from the Toronto Paper Matching System. ACs then bid on these, indi- cating their level of expertise. Based on these bids, and conflicts of interest, approxi- mately 40 papers were assigned to each AC. The ACs then suggested seven reviewers from the pool of Program Committee members for each paper, in ranked order, from which three were chosen automatically by CMT (Microsofts Academic Conference ManagementService),takingloadbalancingandconflictsofinterestintoaccount. The initial reviewing period was five weeks long, after which reviewers provided reviews with preliminary recommendations. With the generous help of several last- minutereviewers,eachpaperreceivedthreereviews.Submissionswithallthreereviews suggesting rejection were independently checked by two ACs and if they agreed, the manuscriptwasrejectedatthisstage(“earlyrejects”).Intotal,334manuscripts(22.5%) were early-rejected, reducing the average AC load toabout 30. Authors of the remaining submissions were then given the opportunity to rebut the reviews, primarily to identify factual errors. Following this, reviewers and ACs dis- cussed papers at length, after which reviewers finalized their reviews and gave a final recommendation to the ACs. Each manuscript was evaluated independently by two ACswhowerenotawareofeachothers,identities.Inmostofthecases,afterextensive discussions,thetwoACsarrivedatacommon decision, whichwasalways adheredto bythePCs.Intheveryfewborderlinecaseswhereanagreementwasnotreached,the PCs acted as tie-breakers. Owing to the rapid expansion of the field, which led to an unexpectedlylargeincreaseinthenumberofsubmissions,thesizeofthevenuebecame a limiting factor and a hard upper bound on the number of accepted papers had to be imposed. We were able to increase the limit by replacing one oral session by a poster session. Nevertheless, this forced the PCs to reject some borderline papers that could otherwise have been accepted. WewanttothankeveryoneinvolvedinmakingtheECCV2016possible.Firstand foremost,thesuccessofECCV2016dependedonthequalityofpaperssubmittedbythe authors,andontheveryhardworkoftheACs,theProgramCommitteemembers,and theadditional reviewers.Weareparticularly gratefultoReneVidalfor his continuous supportandsharingexperiencefromorganizingICCV2015,toLaurentCharlinforthe useoftheTorontoPaperMatchingSystem,toAriKobrenfortheuseoftheResearcher.cc tools,totheComputerVisionFoundation(CVF)forfacilitatingtheuseoftheiThenticate plagiarismdetectionsoftware,andtoGloriaZenandRadu-LaurentiuVieriuforsettingup CMTandmanagingthevarioustoolsinvolved.Wealsooweadebtofgratitudeforthe supportoftheAmsterdamlocalorganizers,especiallyHamdiDibekliogluforkeepingthe Preface IX websitealwaysuptodate.Finally,thepreparationoftheseproceedingswouldnothave been possible without the diligent effort of the publication chairs, Albert Ali Salah and RobbyTan,andofAnnaKramerfromSpringer. October 2016 Bastian Leibe Jiri Matas Nicu Sebe Max Welling Organization General Chairs Theo Gevers University of Amsterdam, The Netherlands Arnold Smeulders University of Amsterdam, The Netherlands Program Committee Co-chairs Bastian Leibe RWTH Aachen, Germany Jiri Matas Czech Technical University, Czech Republic Nicu Sebe University of Trento, Italy Max Welling University of Amsterdam, The Netherlands Honorary Chair Jan Koenderink Delft University of Technology, The Netherlands and KU Leuven, Belgium Advisory Program Chair Luc van Gool ETH Zurich, Switzerland Advisory Workshop Chair Josef Kittler University of Surrey, UK Advisory Conference Chair Alberto del Bimbo University of Florence, Italy Local Arrangements Chairs Hamdi Dibeklioglu Delft University of Technology, The Netherlands Efstratios Gavves University of Amsterdam, The Netherlands Jan van Gemert Delft University of Technology, The Netherlands Thomas Mensink University of Amsterdam, The Netherlands Michael Wilkinson University of Groningen, The Netherlands