Computer Vision ECCV2018 Part VI

所需积分/C币:9 2019-05-20 10:49:54 149.26MB PDF
收藏 收藏

Computer Vision –ECCV 201815th European Conference Munich, Germany, September 8–14, 2018 Proceedings, Part VI。
Moreinformationaboutthisseriesat Ⅴ ittorio ferrari· Martial hebert Cristian Sminchisescu. Yair Weiss(eds) Computer vision ECCV 2018 15th European Conference Munich, Germany, September 8-14, 2018 Proceedings,PartⅥI pringer editors Ⅴ ittorio ferrari Cristian sminchisescu Google research Google research Zurich Zurich Switzerland Switzerland Martial hebert Yair w Carnegie mellon University Hebrew University of Jerusalem Pittsburgh. pa Jerusalem USA ISSN0302-9743 issn 1611-3349(electronic) Lecture Notes in Computer Science ISBN978-3-030-012304 ISBN978-3-030-01231-1( eBook) Library of Congress Control Number: 2018955489 LNCS Sublibrary: SL6-Image Processing, Computer Vision, Pattern Recognition, and graphics O Springer Nature Switzerland AG 2018 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the 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 storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editor give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse ll, 6330 Cham, Switzerland Foreword It was our great pleasure to host the European Conference on Computer Vision 2018 in Munich, Germany. This constituted by far the largest eccv event ever. with close to 2, 900 registered participants and another 600 on the waiting list one month before the conference, participation more than doubled since the last ECCV in Amsterdam. W believe that this is due to a dramatic growth of the computer vision community combined with the popularity of Munich as a major European hub of culture, science, and industry. The conference took place in the heart of Munich in the concert hall Gasteig with workshops and tutorials held at the downtown campus of the Technical University of munich One of the major innovations for ECCV 2018 was the free perpetual availability of all conference and workshop papers, which is often referred to as open access. We note that this is not precisely the same use of the term as in the budapest declaration. Since 2013, CVPR and iCCV have had their papers hosted by the Computer Vision Foun dation(CVF), in parallel with the IEEE Xplore version. This has proved highly ben eficial to the computer vision community. We are delighted to announce that for eccv 2018 a very similar arrangement was put in place with the cooperation of Springer. In particular, the authors final version will be freely available in perpetuity on a CVF page, while Springerlink will continue to host a version with further improvements such as activating reference links and including video. We believe that this will give readers the best of both worlds researchers who are focused on the technical content will have a freely available version in an easily accessible place, while subscribers to SpringerLink will continue to have the additional benefits that this provides. We thank alfred Hofmann from Springer for helping to negotiate this agreement, which we expect will continue for future versions of eccv September 2018 Horst bischof Daniel cremers Bernt schiele Ramin zabih Preface Welcome to the proceedings of the 2018 European Conference on Computer Vision (ECCV 2018) held in Munich, Germany. We are delighted to present this volume reflecting a strong and exciting program, the result of an extensive review process. In total, we received 2, 439 valid paper submissions. Of these, 776 were accepted (31.8%) 717 as posters(29.4%)and 59 as oral presentations(2. 4%0). All oral presentations were presented as posters as well. The program selection process was complicated this year by the large increase in the number of submitted papers, +65% over ECCV 2016, and the use of cmt3 for the first time for a computer vision conference. The program selection process was supported by four program co-chairs(PCs), 126 area chairs (ACS), and 1, 199 reviewers with reviews assigned We were primarily responsible for the design and execution of the review process Beyond administrative rejections, we were involved in acceptance decisions only in the very few cases where the aCs were not able to agree on a decision As PCs, and as is customary in the field, we were not allowed to co-author a submission general co-chairs and other co-organizers who played no role in the review process were permitted to submit papers, and were treated as any other author is Acceptance decisions were made by two independent ACs. The ACs also made a joint recommendation for promoting papers to oral status. We decided on the final selection of oral presentations based on the acs recommendations. There were 126 ACS, selected according to their technical expertise, experience, and geographical diversity(63 from European, nine from Asian/Australian, and 54 from North american institutions). Indeed, 126 ACs is a substantial increase in the number of acs due to the natural increase in the number of papers and to our desire to maintain the number of papers assigned to each ac to a manageable number so as to ensure quality. The acs were aided by the 1, 199 reviewers to whom papers were assigned for reviewing. The Program Committee was selected from committees of previous eccv, iccv, and CVPR conferences and was extended on the basis of suggestions from the ACs Havin a large pool of Program Committee members for reviewing allowed us to match expertise while reducing reviewer loads. No more than eight papers were assigned to a reviewer, maintaining the reviewers'load at the same level as eccv 2016 despite the increase in the number of submitted papers Conflicts of interest between ACs, Program Committee members, and papers were identified based on the home institutions, and on previous collaborations of all researchers involved. To find institutional conflicts all authors Program committee members. and acs were asked to list the internet domains of their current institutions We assigned on average approximately 18 papers to each AC. The papers were assigned using the affinity scores from the Toronto Paper Matching System (TPMs and additional data from the Open Review system, managed by a UMass group. Open Review used additional information from ACs' and authors'records to identify collaborations and to generate matches. Openreview was invaluable in VIII P trace refining conflict definitions and in generating quality matches. The only glitch is that, once the matches were generated, a small percentage of papers were unassigned because of discrepancies between the Open review conflicts and the conflicts entered in CMT3. We manually assigned these papers. This glitch is revealing of the challenge of using multiple systems at once(CMT3 and Open Review in this case), which needs to be addressed in future After assignment of papers to ACS, the ACs suggested seven reviewers per paper from the Program Committee pool. The selection and rank ordering were facilitated by the TPMs affinity scores visible to the ACs for each paper/reviewer pair. The final assignment of papers to reviewers was generated again through Open Review in order to account for refined conflict definitions. This required new features in the Openre view matching system to accommodate the eccv workflow, in particular to incor porate selection ranking, and maximum reviewer load. Very few papers received fewer than three reviewers after matching and were handled through manual assignment Reviewers were then asked to comment on the merit of each paper and to make an initial recommendation ranging from definitely reject to definitely accept, including a borderline rating. The reviewers were also asked to suggest explicit questions they wanted to see answered in the authors'rebuttal. The initial review period was five weeks. Because of the delay in getting all the reviews in, we had to delay the final release of the reviews by four days. however because of the slack included at the tail end of the schedule we were able to maintain the decision target date with sufficient time for all the phases. We reassigned over 100 reviews from 40 reviewers during the review period Unfortunately, the main reason for these reassignments was reviewers declining to review, after having accepted to do so. Other reasons included technical relevance and occasional unidentified conflicts. We express our thanks to the emer gency reviewers who generously accepted to perform these reviews under short notice In addition a substantial number of manual corrections had to do with reviewers using a different email address than the one that was used at the time of the reviewer invitation. This is revealing of a broader issue with identifying users by email addresses that change frequently enough to cause significant problems during the timespan of the conference process The authors were then given the opportunity to rebut the reviews, to identify factual errors, and to address the specific questions raised by the reviewers over a seven-day rebuttal period. The exact format of the rebuttal was the object of considerable debate among the organizers, as well as with prior organizers. At issue is to balance giving the author the opportunity to respond completely and precisely to the reviewers, e. g, by including graphs of experiments, while avoiding requests for completely new material or experimental results not included in the original paper. In the end, we decided on the two-page PDF document in conference format. Following this rebuttal period, reviewers and ACs discussed papers at length, after which reviewers finalized their evaluation and gave a final recommendation to the ACs. A significant percentage of the reviewers did enter their final recommendation if it did not differ from their initial recommendation. Given the tight schedule, we did not wait until all were entered After this discussion period, each paper was assigned to a second AC. The AC/paper matching was again run through OpenReview. Again, the Open review team worked quickly to implement the features specific to this process, in this case accounting for the Preface IX existing AC assignment, as well as minimizing the fragmentation across ACs, so that each AC had on average only 5.5 buddy acs to communicate with The largest number was 11. Given the complexity of the conflicts, this was a very efficient set of assign- ments from Open Review. Each paper was then evaluated by its assigned pair of acs For each paper, we required each of the two ACs assigned to certify both the final recommendation and the metareview (aka consolidation report). In all cases, after extensive discussions, the two aCs arrived at a common acceptance decision we maintained these decisions, with the caveat that we did evaluate sometimes going back to the ACs, a few papers for which the final acceptance decision substantially deviated from the consensus from the reviewers, amending three decisions in the process We want to thank everyone involved in making eccv 2018 possible. The success of eCCv 2018 depended on the quality of papers submitted by the authors, and on the very hard work of the acs and the program committee members We are particularly grateful to the Open review team (Melisa bok, Ari Kobren, Andrew mccallum Michael Spector) for their support, in particular their willingness to implement new features, often on a tight schedule, to laurent Charlin for the use of the toronto paper Matching System, to the CMt3 team, in particular in dealing with all the issues that arise when using a new system, to Friedrich Fraundorfer and Quirin Lohr for main taining the online version of the program, and to the cmu staff (Keyla Cook, Lynnetta Miller, Ashley Song, Nora Kazour) for assisting with data entry/editing in CMT3 Finally, the preparation of these proceedings would not have been possible without the diligent effort of the publication chairs, Albert Ali Salah and Hamdi Dibeklioglu, and of Anna Kramer and Alfred Hofmann from Springer September 2018 Vittorio ferrari Cristian sminchisescu Yair Weiss Organization General chairs Horst bischof Graz University of Technology, Austria Daniel cremers Technical University of Munich, Germany Bernt schiele Saarland University, Max Planck Institute for Informatics Germany Ramin zabih CornellNYctech USA Program Committee Co-chairs Ⅴ ittorio ferrari University of Edinburgh, UK Martial hebert Carnegie Mellon University, USA Cristian sminchisescu Lund university, Sweden Yair Weiss Hebrew University, Israe Local Arrangements Chairs Bjorn menze Technical University of Munich, Germany Matthias niessner Technical University of Munich, Germany Workshop Chairs Stefan roth TU Darmstadt, Germany Laura Leal-taixe Technical University of Munich, Germany utorial chairs Michael bronstein Universita della svizzera italiana. switzerland Laura leal-taixe Technical University of Munich, Germany Website chair Friedrich fraundorfer Graz University of Technology, Austria Demo chairs Federico tombari Technical University of Munich, Germany Joerg Steckler Technical University of Munich, Germany

试读 127P Computer Vision ECCV2018 Part VI
限时抽奖 低至0.43元/次
身份认证后 购VIP低至7折
  • 分享王者

关注 私信
Computer Vision ECCV2018 Part VI 9积分/C币 立即下载
Computer Vision ECCV2018 Part VI第1页
Computer Vision ECCV2018 Part VI第2页
Computer Vision ECCV2018 Part VI第3页
Computer Vision ECCV2018 Part VI第4页
Computer Vision ECCV2018 Part VI第5页
Computer Vision ECCV2018 Part VI第6页
Computer Vision ECCV2018 Part VI第7页
Computer Vision ECCV2018 Part VI第8页
Computer Vision ECCV2018 Part VI第9页
Computer Vision ECCV2018 Part VI第10页
Computer Vision ECCV2018 Part VI第11页
Computer Vision ECCV2018 Part VI第12页
Computer Vision ECCV2018 Part VI第13页
Computer Vision ECCV2018 Part VI第14页
Computer Vision ECCV2018 Part VI第15页
Computer Vision ECCV2018 Part VI第16页
Computer Vision ECCV2018 Part VI第17页
Computer Vision ECCV2018 Part VI第18页
Computer Vision ECCV2018 Part VI第19页
Computer Vision ECCV2018 Part VI第20页

试读结束, 可继续阅读

9积分/C币 立即下载