David Fleet Tomas Pajdla Bernt Schiele Tinne Tuytelaars (Eds.) 2 Computer Vision – 9 6 8 S ECCV 2014 C N L 13th European Conference Zurich, Switzerland, September 6–12, 2014 Proceedings, Part IV 123 Lecture Notes in Computer Science 8692 CommencedPublicationin1973 FoundingandFormerSeriesEditors: GerhardGoos,JurisHartmanis,andJanvanLeeuwen EditorialBoard DavidHutchison LancasterUniversity,UK TakeoKanade CarnegieMellonUniversity,Pittsburgh,PA,USA JosefKittler UniversityofSurrey,Guildford,UK JonM.Kleinberg CornellUniversity,Ithaca,NY,USA AlfredKobsa UniversityofCalifornia,Irvine,CA,USA FriedemannMattern ETHZurich,Switzerland JohnC.Mitchell StanfordUniversity,CA,USA MoniNaor WeizmannInstituteofScience,Rehovot,Israel OscarNierstrasz UniversityofBern,Switzerland C.PanduRangan IndianInstituteofTechnology,Madras,India BernhardSteffen TUDortmundUniversity,Germany DemetriTerzopoulos UniversityofCalifornia,LosAngeles,CA,USA DougTygar UniversityofCalifornia,Berkeley,CA,USA GerhardWeikum MaxPlanckInstituteforInformatics,Saarbruecken,Germany David Fleet Tomas Pajdla Bernt Schiele Tinne Tuytelaars (Eds.) Computer Vision – ECCV 2014 13th European Conference Zurich, Switzerland, September 6-12, 2014 Proceedings, Part IV 1 3 VolumeEditors DavidFleet UniversityofToronto,DepartmentofComputerScience 6King’sCollegeRoad,Toronto,ONM5H3S5,Canada E-mail:fl[email protected] TomasPajdla CzechTechnicalUniversityinPrague,DepartmentofCybernetics Technicka2,16627Prague6,CzechRepublic E-mail:[email protected] BerntSchiele Max-Planck-InstitutfürInformatik CampusE14,66123Saarbrücken,Germany E-mail:[email protected] TinneTuytelaars KULeuven,ESAT-PSI,iMinds KasteelparkArenberg10,Bus2441,3001Leuven,Belgium E-mail:[email protected] Videostothisbookcanbeaccessedat http://www.springerimages.com/videos/978-3-319-10592-5 ISSN0302-9743 e-ISSN1611-3349 ISBN978-3-319-10592-5 e-ISBN978-3-319-10593-2 DOI10.1007/978-3-319-10593-2 SpringerChamHeidelbergNewYorkDordrechtLondon LibraryofCongressControlNumber:2014946360 LNCSSublibrary:SL6–ImageProcessing,ComputerVision,PatternRecognition, andGraphics ©SpringerInternationalPublishingSwitzerland2014 Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation, broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology nowknownorhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerptsinconnection withreviewsorscholarlyanalysisormaterialsuppliedspecificallyforthepurposeofbeingenteredand executedonacomputersystem,forexclusiveusebythepurchaserofthework.Duplicationofthispublication orpartsthereofispermittedonlyundertheprovisionsoftheCopyrightLawofthePublisher’slocation, inistcurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.Permissionsforuse maybeobtainedthroughRightsLinkattheCopyrightClearanceCenter.Violationsareliabletoprosecution undertherespectiveCopyrightLaw. Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse. Whiletheadviceandinformationinthisbookarebelievedtobetrueandaccurateatthedateofpublication, neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityforanyerrorsor omissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,withrespecttothe materialcontainedherein. Typesetting:Camera-readybyauthor,dataconversionbyScientificPublishingServices,Chennai,India Printedonacid-freepaper SpringerispartofSpringerScience+BusinessMedia(www.springer.com) Foreword The European Conference on Computer Vision is one of the top conferences in computer vision. It was first held in 1990 in Antibes (France) with subsequent conferences in Santa Margherita Ligure (Italy) in 1992, Stockholm (Sweden) in 1994,Cambridge(UK)in1996,Freiburg(Germany)in1998,Dublin(Ireland)in 2000, Copenhagen (Denmark) in 2002, Prague (Czech Republic) in 2004, Graz (Austria) in 2006, Marseille (France) in 2008, Heraklion (Greece) in 2010, and Florence(Italy)in2012.Manypeoplehaveworkedhardtoturnthe2014edition into as great a success. We hope you will find this a mission accomplished. The chairs decided to adhere to the classic single-track scheme. In terms of the time ordering, we decided to largely follow the Florence example (typically starting with poster sessions, followed by oral sessions), which offers a lot of flexibility to network and is more forgiving for the not-so-early-birds and hard- core gourmets. A large conference like ECCV requires the help of many. They made sure there was a full program including the main conference, tutorials, workshops, exhibits,demos,proceedings,videostreaming/archive,andWebdescriptions.We wanttocordiallythankallthosevolunteers!Pleasehavealookattheconference website to see their names (http://eccv2014.org/people/). We also thank our generous sponsors. Their support was vital for keeping prices low and enriching the program.And itis goodto seesuchalevel ofindustrial interestin whatour community is doing! We hope you will enjoy the proceedings ECCV 2014. Also, willkommen in Zu¨rich! September 2014 Marc Pollefeys Luc Van Gool General Chairs Preface Welcome to the proceedingsof the 2014EuropeanConference onComputer Vi- sion (ECCV 2014)that was in Zurich, Switzerland. We are delighted to present this volume reflecting a strong and exciting program, the result of an extensive review process. In total, we received 1,444 paper submissions. Of these, 85 vio- latedthe ECCVsubmissionguidelines andwererejectedwithoutreview. Ofthe remainder, 363 were accepted (26,7%): 325 as posters (23,9%) and 38 as oral presentations (2,8%). This selection process was a combined effort of four pro- gram co-chairs (PCs), 53 area chairs (ACs), 803 Program Committee members and 247 additional reviewers. As PCs we were primarily responsible for the design and execution of the review process. Beyond administrative rejections, we were not directly involved inacceptance decisions.Becausethe generalco-chairswerepermitted to submit papers, they played no role in the review process and were treated as any other author. AcceptancedecisionsweremadebytheACCommittee.Therewere53ACsin total, selected by the PCs to provide sufficient technical expertise, geographical diversity (21 from Europe, 7 from Asia, and 25 from North America)and a mix of AC experience (7 had no previous AC experience, 18 had served as AC of a major international vision conference once since 2010, 8 had served twice, 13 had served three times, and 7 had served 4 times). ACs were aided by 803 ProgramCommittee members to whom papers were assignedfor reviewing. There were 247 additional reviewers, each supervised by a Program Committee member. The Program Committee was based on sug- gestions from ACs, and committees from previous conferences. Google Scholar profiles were collected for all candidate Program Committee members and vet- ted by PCs. Having a large pool of ProgramCommittee members for reviewing allowedustomatchexpertisewhileboundingreviewerloads.Nomorethannine paperswereassignedtoanyoneProgramCommitteemember,withamaximum of six to graduate students. The ECCV2014review processwas double blind. Authors did not knowthe reviewers’ identities, nor the ACs handling their paper(s). We did our utmost to ensure that ACs and reviewers did not know authors’ identities, even though anonymity becomes difficult to maintain as more and more submissions appear concurrently on arXiv.org. Particular attention was paid to minimizing potential conflicts of interest. Conflicts of interest between ACs, Program Committee members, and papers were based on authorship of ECCV 2014 submissions, on their home institu- tions, andonpreviouscollaborations.Tofind institutional conflicts, all authors, VIII Preface ProgramCommittee members, andACs wereaskedto list the Internetdomains of their current institutions. To find collaborators, the DBLP (www.dblp.org) database was used to find any co-authored papers in the period 2010–2014. Weinitially assignedapproximately100paperstoeachAC, basedonaffinity scores from the Toronto Paper Matching System and authors’ AC suggestions. ACs then bid on these, indicating their level of expertise. Based on these bids, and conflicts of interest, approximately 27 papers were assigned to each AC, for which they would act as the primary AC. The primary AC then suggested seven reviewers from the pool of Program Committee members (in rank order) for each paper, from which three were chosen per paper, taking load balancing and conflicts of interest into account. Many papers were also assigneda secondaryAC, either directly by the PCs, orasaconsequenceoftheprimaryACrequestingtheaidofanACwithcomple- mentaryexpertise.SecondaryACscouldbeassignedatanystageintheprocess, but in most cases this occurred about two weeks before the final AC meeting. Hence, inadditiontotheir initial loadofapproximately27papers,eachAC was askedtohandlethreetofivemorepapersasasecondaryAC;theywereexpected to read and write a short assessment of such papers. In addition, two of the 53 ACs were not directly assigned papers. Rather, they were available through- out the process to aid other ACs at any stage (e.g., with decisions, evaluating technical issues, additional reviews, etc.). The initial reviewing period was three weeks long, after which reviewers provided reviews with preliminary recommendations. Three weeks is somewhat shorterthannormal,butthisdidnotseemtocauseanyunusualproblems.With the generous help of several last-minute reviewers, each paper received three reviews. Authors were then given the opportunity to rebut the reviews, primarily to identify any factual errors. Following this, reviewers and ACs discussed papers at length, after which reviewers finalized their reviews and gave a final recom- mendation to the ACs. Many ACs requested help from secondary ACs at this time. Papers, for which rejection was clear and certain, based on the reviews and the AC’s assessment, were identified by their primary ACs and vetted by a shadow AC prior to rejection. (These shadow ACs were assigned by the PCs.) All papers with any chance of acceptance were further discussed at the AC meeting. Those deemed“strong”by primaryACs (about140intotal) werealso assigned a secondary AC. The AC meeting, with all but two of the primary ACs present, took place in Zurich.ACsweredividedinto17tripletsforeachmorning,andadifferentsetof tripletsforeachafternoon.Giventhecontentofthethree(ormore)reviewsalong with reviewer recommendations, rebuttals, online discussions among reviewers and primary ACs, written input from and discussions with secondary ACs, the Preface IX ACtripletsthenworkedtogethertoresolvequestions,calibrateassessments,and make acceptance decisions. To select oral presentations, all strong papers, along with any others put forwardbytriplets(about155intotal), werethendiscussedinfourpanels,each comprising four or five triplets. Each panel ranked these oral candidates, using fourcategories.Papersinthe twotopcategoriesprovidedthe finalsetof38oral presentations. We want to thank everyone involved in making the ECCV 2014 Program possible.Firstandforemost,the successofECCV2014dependedonthequality of papers submitted by authors, and on the very hard work of the reviewers, the ProgramCommittee members and the ACs. We are particularly grateful to Kyros Kutulakos for his enormous software support before and during the AC meeting, to LaurentCharlinfor the use ofthe TorontoPaperMatching System, and Chaohui Wang for help optimizing the assignment of papers to ACs. We also owe a debt of gratitude for the great support of Zurich local organizers, especially Susanne Keller and her team. September 2014 David Fleet Tomas Pajdla Bernt Schiele Tinne Tuytelaars Organization General Chairs Luc Van Gool ETH Zurich, Switzerland Marc Pollefeys ETH Zurich, Switzerland Program Chairs Tinne Tuytelaars KU Leuven, Belgium Bernt Schiele MPI Informatics, Saarbru¨cken, Germany Tomas Pajdla CTU Prague, Czech Republic David Fleet University of Toronto, Canada Local Arrangements Chairs Konrad Schindler ETH Zurich, Switzerland Vittorio Ferrari University of Edinburgh, UK Workshop Chairs Lourdes Agapito University College London, UK Carsten Rother TU Dresden, Germany Michael Bronstein University of Lugano, Switzerland Tutorial Chairs Bastian Leibe RWTH Aachen, Germany Paolo Favaro University of Bern, Switzerland Christoph Lampert IST Austria Poster Chair Helmut Grabner ETH Zurich, Switzerland Publication Chairs Mario Fritz MPI Informatics, Saarbru¨cken, Germany Michael Stark MPI Informatics, Saarbru¨cken, Germany XII Organization Demo Chairs Davide Scaramuzza University of Zurich, Switzerland Jan-Michael Frahm University of North Carolina at Chapel Hill, USA Exhibition Chair Tamar Tolcachier University of Zurich, Switzerland Industrial Liaison Chairs Alexander Sorkine-Hornung Disney Research Zurich, Switzerland Fatih Porikli ANU, Australia Student Grant Chair Seon Joo Kim Yonsei University, Korea Air Shelters Accommodation Chair Maros Blaha ETH Zurich, Switzerland Website Chairs Lorenz Meier ETH Zurich, Switzerland Bastien Jacquet ETH Zurich, Switzerland Internet Chair Thorsten Steenbock ETH Zurich, Switzerland Student Volunteer Chairs Andrea Cohen ETH Zurich, Switzerland Ralf Dragon ETH Zurich, Switzerland Laura Leal-Taix´e ETH Zurich, Switzerland Finance Chair Amael Delaunoy ETH Zurich, Switzerland Conference Coordinator Susanne H. Keller ETH Zurich, Switzerland