Computer Vision ECCV 2018 15th European Conference Munich Germany September 8 14 2018 Proceedings Part I Vittorio Ferrari

Computer Vision ECCV 2018 15th
European Conference Munich Germany

September 8 14 2018 Proceedings Part I
Vittorio Ferrari
Visit to download the full and correct content document:
https://textbookfull.com/product/computer-vision-eccv-2018-15th-european-conferenc
e-munich-germany-september-8-14-2018-proceedings-part-i-vittorio-ferrari/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Computer Vision ECCV 2018 15th European Conference

Munich Germany September 8 14 2018 Proceedings Part III
Vittorio Ferrari
https://textbookfull.com/product/computer-vision-eccv-2018-15th-
european-conference-munich-germany-
september-8-14-2018-proceedings-part-iii-vittorio-ferrari/

Munich Germany September 8 14 2018 Proceedings Part IV
Vittorio Ferrari
september-8-14-2018-proceedings-part-iv-vittorio-ferrari/

Munich Germany September 8 14 2018 Proceedings Part VII
Vittorio Ferrari
september-8-14-2018-proceedings-part-vii-vittorio-ferrari/

Munich Germany September 8 14 2018 Proceedings Part X
Vittorio Ferrari
september-8-14-2018-proceedings-part-x-vittorio-ferrari/
Munich Germany September 8 14 2018 Proceedings Part
XIII Vittorio Ferrari
september-8-14-2018-proceedings-part-xiii-vittorio-ferrari/

Munich Germany September 8 14 2018 Proceedings Part XIV
Vittorio Ferrari
september-8-14-2018-proceedings-part-xiv-vittorio-ferrari/
Computer Vision ECCV 2018 Workshops Munich Germany

September 8 14 2018 Proceedings Part II Laura Leal-
Taixé
https://textbookfull.com/product/computer-vision-
eccv-2018-workshops-munich-germany-
september-8-14-2018-proceedings-part-ii-laura-leal-taixe/
Computer Vision ECCV 2018 Workshops Munich Germany

September 8 14 2018 Proceedings Part V Laura Leal-Taixé
https://textbookfull.com/product/computer-vision-
eccv-2018-workshops-munich-germany-
september-8-14-2018-proceedings-part-v-laura-leal-taixe/

Amsterdam The Netherlands October 11 14 2016
Proceedings Part I 1st Edition Bastian Leibe
european-conference-amsterdam-the-netherlands-
october-11-14-2016-proceedings-part-i-1st-edition-bastian-leibe/
Vittorio Ferrari · Martial Hebert
Cristian Sminchisescu
Yair Weiss (Eds.)
LNCS 11205
Computer Vision –
ECCV 2018
15th European Conference
Munich, Germany, September 8–14, 2018
Proceedings, Part I
123
Lecture Notes in Computer Science 11205
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, Chennai, 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
Vittorio Ferrari Martial Hebert
•
Cristian Sminchisescu Yair Weiss (Eds.)

•
Computer Vision –
ECCV 2018
15th European Conference
Munich, Germany, September 8–14, 2018
Proceedings, Part I
123
Editors
Vittorio Ferrari Cristian Sminchisescu
Google Research Google Research
Zurich Zurich
Switzerland Switzerland
Martial Hebert Yair Weiss
Carnegie Mellon University Hebrew University of Jerusalem
Pittsburgh, PA Jerusalem
USA Israel
ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-030-01245-8 ISBN 978-3-030-01246-5 (eBook)
https://doi.org/10.1007/978-3-030-01246-5
Library of Congress Control Number: 2018955489
LNCS Sublibrary: SL6 – Image Processing, Computer Vision, Pattern Recognition, and Graphics
© 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 editors
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 11, 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. We
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 author’s 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%). 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. Having
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 OpenReview system, managed by a UMass
group. OpenReview used additional information from ACs’ and authors’ records to
identify collaborations and to generate matches. OpenReview was invaluable in
VIII Preface
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 OpenReview 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 OpenReview 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 OpenReview 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 OpenReview 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 OpenReview. 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 OpenReview 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 Dibeklioğlu, and of
Anna Kramer and Alfred Hofmann from Springer.
September 2018 Vittorio Ferrari

Martial Hebert
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

Vittorio Ferrari University of Edinburgh, UK
Martial Hebert Carnegie Mellon University, USA
Cristian Sminchisescu Lund University, Sweden
Yair Weiss Hebrew University, Israel
Local Arrangements Chairs

Björn Menze Technical University of Munich, Germany
Matthias Niessner Technical University of Munich, Germany
Workshop Chairs
Stefan Roth TU Darmstadt, Germany
Laura Leal-Taixé Technical University of Munich, Germany
Tutorial Chairs
Michael Bronstein Università della Svizzera Italiana, Switzerland
Website Chair
Friedrich Fraundorfer Graz University of Technology, Austria
Demo Chairs
Federico Tombari Technical University of Munich, Germany
Joerg Stueckler Technical University of Munich, Germany
XII Organization
Publicity Chair
Giovanni Maria University of Catania, Italy
Farinella
Industrial Liaison Chairs

Florent Perronnin Naver Labs, France
Yunchao Gong Snap, USA
Helmut Grabner Logitech, Switzerland
Finance Chair
Gerard Medioni Amazon, University of Southern California, USA
Publication Chairs
Albert Ali Salah Boğaziçi University, Turkey
Hamdi Dibeklioğlu Bilkent University, Turkey
Area Chairs
Kalle Åström Lund University, Sweden
Zeynep Akata University of Amsterdam, The Netherlands
Joao Barreto University of Coimbra, Portugal
Ronen Basri Weizmann Institute of Science, Israel
Dhruv Batra Georgia Tech and Facebook AI Research, USA
Serge Belongie Cornell University, USA
Rodrigo Benenson Google, Switzerland
Hakan Bilen University of Edinburgh, UK
Matthew Blaschko KU Leuven, Belgium
Edmond Boyer Inria, France
Gabriel Brostow University College London, UK
Thomas Brox University of Freiburg, Germany
Marcus Brubaker York University, Canada
Barbara Caputo Politecnico di Torino and the Italian Institute
of Technology, Italy
Tim Cootes University of Manchester, UK
Trevor Darrell University of California, Berkeley, USA
Larry Davis University of Maryland at College Park, USA
Andrew Davison Imperial College London, UK
Fernando de la Torre Carnegie Mellon University, USA
Irfan Essa GeorgiaTech, USA
Ali Farhadi University of Washington, USA
Paolo Favaro University of Bern, Switzerland
Michael Felsberg Linköping University, Sweden
Organization XIII
Sanja Fidler University of Toronto, Canada

Andrew Fitzgibbon Microsoft, Cambridge, UK
David Forsyth University of Illinois at Urbana-Champaign, USA
Charless Fowlkes University of California, Irvine, USA
Bill Freeman MIT, USA
Mario Fritz MPII, Germany
Jürgen Gall University of Bonn, Germany
Dariu Gavrila TU Delft, The Netherlands
Andreas Geiger MPI-IS and University of Tübingen, Germany
Theo Gevers University of Amsterdam, The Netherlands
Ross Girshick Facebook AI Research, USA
Kristen Grauman Facebook AI Research and UT Austin, USA
Abhinav Gupta Carnegie Mellon University, USA
Kaiming He Facebook AI Research, USA
Martial Hebert Carnegie Mellon University, USA
Anders Heyden Lund University, Sweden
Timothy Hospedales University of Edinburgh, UK
Michal Irani Weizmann Institute of Science, Israel
Phillip Isola University of California, Berkeley, USA
Hervé Jégou Facebook AI Research, France
David Jacobs University of Maryland, College Park, USA
Allan Jepson University of Toronto, Canada
Jiaya Jia Chinese University of Hong Kong, SAR China
Fredrik Kahl Chalmers University, USA
Hedvig Kjellström KTH Royal Institute of Technology, Sweden
Iasonas Kokkinos University College London and Facebook, UK
Vladlen Koltun Intel Labs, USA
Philipp Krähenbühl UT Austin, USA
M. Pawan Kumar University of Oxford, UK
Kyros Kutulakos University of Toronto, Canada
In Kweon KAIST, South Korea
Ivan Laptev Inria, France
Svetlana Lazebnik University of Illinois at Urbana-Champaign, USA
Erik Learned-Miller University of Massachusetts, Amherst, USA
Kyoung Mu Lee Seoul National University, South Korea
Bastian Leibe RWTH Aachen University, Germany
Aleš Leonardis University of Birmingham, UK
Vincent Lepetit University of Bordeaux, France and Graz University
of Technology, Austria
Fuxin Li Oregon State University, USA
Dahua Lin Chinese University of Hong Kong, SAR China
Jim Little University of British Columbia, Canada
Ce Liu Google, USA
Chen Change Loy Nanyang Technological University, Singapore
Jiri Matas Czech Technical University in Prague, Czechia
XIV Organization
Yasuyuki Matsushita Osaka University, Japan

Dimitris Metaxas Rutgers University, USA
Greg Mori Simon Fraser University, Canada
Vittorio Murino Istituto Italiano di Tecnologia, Italy
Richard Newcombe Oculus Research, USA
Minh Hoai Nguyen Stony Brook University, USA
Sebastian Nowozin Microsoft Research Cambridge, UK
Aude Oliva MIT, USA
Bjorn Ommer Heidelberg University, Germany
Tomas Pajdla Czech Technical University in Prague, Czechia
Maja Pantic Imperial College London and Samsung AI Research
Centre Cambridge, UK
Caroline Pantofaru Google, USA
Devi Parikh Georgia Tech and Facebook AI Research, USA
Sylvain Paris Adobe Research, USA
Vladimir Pavlovic Rutgers University, USA
Marcello Pelillo University of Venice, Italy
Patrick Pérez Valeo, France
Robert Pless George Washington University, USA
Thomas Pock Graz University of Technology, Austria
Jean Ponce Inria, France
Gerard Pons-Moll MPII, Saarland Informatics Campus, Germany
Long Quan Hong Kong University of Science and Technology,
SAR China
Stefan Roth TU Darmstadt, Germany
Carsten Rother University of Heidelberg, Germany
Bryan Russell Adobe Research, USA
Kate Saenko Boston University, USA
Mathieu Salzmann EPFL, Switzerland
Dimitris Samaras Stony Brook University, USA
Yoichi Sato University of Tokyo, Japan
Silvio Savarese Stanford University, USA
Konrad Schindler ETH Zurich, Switzerland
Cordelia Schmid Inria, France and Google, France
Nicu Sebe University of Trento, Italy
Fei Sha University of Southern California, USA
Greg Shakhnarovich TTI Chicago, USA
Jianbo Shi University of Pennsylvania, USA
Abhinav Shrivastava UMD and Google, USA
Yan Shuicheng National University of Singapore, Singapore
Leonid Sigal University of British Columbia, Canada
Josef Sivic Czech Technical University in Prague, Czechia
Arnold Smeulders University of Amsterdam, The Netherlands
Deqing Sun NVIDIA, USA
Antonio Torralba MIT, USA
Zhuowen Tu University of California, San Diego, USA
Organization XV
Tinne Tuytelaars KU Leuven, Belgium

Jasper Uijlings Google, Switzerland
Joost van de Weijer Computer Vision Center, Spain
Nuno Vasconcelos University of California, San Diego, USA
Andrea Vedaldi University of Oxford, UK
Olga Veksler University of Western Ontario, Canada
Jakob Verbeek Inria, France
Rene Vidal Johns Hopkins University, USA
Daphna Weinshall Hebrew University, Israel
Chris Williams University of Edinburgh, UK
Lior Wolf Tel Aviv University, Israel
Ming-Hsuan Yang University of California at Merced, USA
Todd Zickler Harvard University, USA
Andrew Zisserman University of Oxford, UK
Technical Program Committee
Hassan Abu Alhaija Peter Anderson Arunava Banerjee

Radhakrishna Achanta Juan Andrade-Cetto Atsuhiko Banno
Hanno Ackermann Mykhaylo Andriluka Aayush Bansal
Ehsan Adeli Anelia Angelova Yingze Bao
Lourdes Agapito Michel Antunes Md Jawadul Bappy
Aishwarya Agrawal Pablo Arbelaez Pierre Baqué
Antonio Agudo Vasileios Argyriou Dániel Baráth
Eirikur Agustsson Chetan Arora Adrian Barbu
Karim Ahmed Federica Arrigoni Kobus Barnard
Byeongjoo Ahn Vassilis Athitsos Nick Barnes
Unaiza Ahsan Mathieu Aubry Francisco Barranco
Emre Akbaş Shai Avidan Adrien Bartoli
Eren Aksoy Yannis Avrithis E. Bayro-Corrochano
Yağız Aksoy Samaneh Azadi Paul Beardlsey
Alexandre Alahi Hossein Azizpour Vasileios Belagiannis
Jean-Baptiste Alayrac Artem Babenko Sean Bell
Samuel Albanie Timur Bagautdinov Ismail Ben
Cenek Albl Andrew Bagdanov Boulbaba Ben Amor
Saad Ali Hessam Bagherinezhad Gil Ben-Artzi
Rahaf Aljundi Yuval Bahat Ohad Ben-Shahar
Jose M. Alvarez Min Bai Abhijit Bendale
Humam Alwassel Qinxun Bai Rodrigo Benenson
Toshiyuki Amano Song Bai Fabian Benitez-Quiroz
Mitsuru Ambai Xiang Bai Fethallah Benmansour
Mohamed Amer Peter Bajcsy Ryad Benosman
Senjian An Amr Bakry Filippo Bergamasco
Cosmin Ancuti Kavita Bala David Bermudez
XVI Organization
Jesus Bermudez-Cameo Xun Cao Erkang Cheng

Leonard Berrada Yanshuai Cao Jingchun Cheng
Gedas Bertasius Joao Carreira Ming-Ming Cheng
Ross Beveridge Dan Casas Wen-Huang Cheng
Lucas Beyer Daniel Castro Yuan Cheng
Bir Bhanu Jan Cech Anoop Cherian
S. Bhattacharya M. Emre Celebi Liang-Tien Chia
Binod Bhattarai Duygu Ceylan Naoki Chiba
Arnav Bhavsar Menglei Chai Shao-Yi Chien
Simone Bianco Ayan Chakrabarti Han-Pang Chiu
Adel Bibi Rudrasis Chakraborty Wei-Chen Chiu
Pia Bideau Shayok Chakraborty Nam Ik Cho
Josef Bigun Tat-Jen Cham Sunghyun Cho
Arijit Biswas Antonin Chambolle TaeEun Choe
Soma Biswas Antoni Chan Jongmoo Choi
Marten Bjoerkman Sharat Chandran Christopher Choy
Volker Blanz Hyun Sung Chang Wen-Sheng Chu
Vishnu Boddeti Ju Yong Chang Yung-Yu Chuang
Piotr Bojanowski Xiaojun Chang Ondrej Chum
Terrance Boult Soravit Changpinyo Joon Son Chung
Yuri Boykov Wei-Lun Chao Gökberk Cinbis
Hakan Boyraz Yu-Wei Chao James Clark
Eric Brachmann Visesh Chari Andrea Cohen
Samarth Brahmbhatt Rizwan Chaudhry Forrester Cole
Mathieu Bredif Siddhartha Chaudhuri Toby Collins
Francois Bremond Rama Chellappa John Collomosse
Michael Brown Chao Chen Camille Couprie
Luc Brun Chen Chen David Crandall
Shyamal Buch Cheng Chen Marco Cristani
Pradeep Buddharaju Chu-Song Chen Canton Cristian
Aurelie Bugeau Guang Chen James Crowley
Rudy Bunel Hsin-I Chen Yin Cui
Xavier Burgos Artizzu Hwann-Tzong Chen Zhaopeng Cui
Darius Burschka Kai Chen Bo Dai
Andrei Bursuc Kan Chen Jifeng Dai
Zoya Bylinskii Kevin Chen Qieyun Dai
Fabian Caba Liang-Chieh Chen Shengyang Dai
Daniel Cabrini Hauagge Lin Chen Yuchao Dai
Cesar Cadena Lerma Qifeng Chen Carlo Dal Mutto
Holger Caesar Ting Chen Dima Damen
Jianfei Cai Wei Chen Zachary Daniels
Junjie Cai Xi Chen Kostas Daniilidis
Zhaowei Cai Xilin Chen Donald Dansereau
Simone Calderara Xinlei Chen Mohamed Daoudi
Neill Campbell Yingcong Chen Abhishek Das
Octavia Camps Yixin Chen Samyak Datta
Organization XVII
Achal Dave Aykut Erdem Ryo Furukawa

Shalini De Mello Erkut Erdem Yasutaka Furukawa
Teofilo deCampos Hugo Jair Escalante Andrea Fusiello
Joseph DeGol Sergio Escalera Fatma Güney
Koichiro Deguchi Victor Escorcia Raghudeep Gadde
Alessio Del Bue Francisco Estrada Silvano Galliani
Stefanie Demirci Davide Eynard Orazio Gallo
Jia Deng Bin Fan Chuang Gan
Zhiwei Deng Jialue Fan Bin-Bin Gao
Joachim Denzler Quanfu Fan Jin Gao
Konstantinos Derpanis Chen Fang Junbin Gao
Aditya Deshpande Tian Fang Ruohan Gao
Alban Desmaison Yi Fang Shenghua Gao
Frédéric Devernay Hany Farid Animesh Garg
Abhinav Dhall Giovanni Farinella Ravi Garg
Michel Dhome Ryan Farrell Erik Gartner
Hamdi Dibeklioğlu Alireza Fathi Simone Gasparin
Mert Dikmen Christoph Feichtenhofer Jochen Gast
Cosimo Distante Wenxin Feng Leon A. Gatys
Ajay Divakaran Martin Fergie Stratis Gavves
Mandar Dixit Cornelia Fermuller Liuhao Ge
Carl Doersch Basura Fernando Timnit Gebru
Piotr Dollar Michael Firman James Gee
Bo Dong Bob Fisher Peter Gehler
Chao Dong John Fisher Xin Geng
Huang Dong Mathew Fisher Guido Gerig
Jian Dong Boris Flach David Geronimo
Jiangxin Dong Matt Flagg Bernard Ghanem
Weisheng Dong Francois Fleuret Michael Gharbi
Simon Donné David Fofi Golnaz Ghiasi
Gianfranco Doretto Ruth Fong Spyros Gidaris
Alexey Dosovitskiy Gian Luca Foresti Andrew Gilbert
Matthijs Douze Per-Erik Forssén Rohit Girdhar
Bruce Draper David Fouhey Ioannis Gkioulekas
Bertram Drost Katerina Fragkiadaki Georgia Gkioxari
Liang Du Victor Fragoso Guy Godin
Shichuan Du Jan-Michael Frahm Roland Goecke
Gregory Dudek Jean-Sebastien Franco Michael Goesele
Zoran Duric Ohad Fried Nuno Goncalves
Pınar Duygulu Simone Frintrop Boqing Gong
Hazım Ekenel Huazhu Fu Minglun Gong
Tarek El-Gaaly Yun Fu Yunchao Gong
Ehsan Elhamifar Olac Fuentes Abel Gonzalez-Garcia
Mohamed Elhoseiny Christopher Funk Daniel Gordon
Sabu Emmanuel Thomas Funkhouser Paulo Gotardo
Ian Endres Brian Funt Stephen Gould
XVIII Organization
Venu Govindu Wolfgang Heidrich Evren Imre

Helmut Grabner Janne Heikkila Eldar Insafutdinov
Petr Gronat Jared Heinly Go Irie
Steve Gu Mattias Heinrich Hossam Isack
Josechu Guerrero Lisa Anne Hendricks Ahmet Işcen
Anupam Guha Dan Hendrycks Daisuke Iwai
Jean-Yves Guillemaut Stephane Herbin Hamid Izadinia
Alp Güler Alexander Hermans Nathan Jacobs
Erhan Gündoğdu Luis Herranz Suyog Jain
Guodong Guo Aaron Hertzmann Varun Jampani
Xinqing Guo Adrian Hilton C. V. Jawahar
Ankush Gupta Michael Hirsch Dinesh Jayaraman
Mohit Gupta Steven Hoi Sadeep Jayasumana
Saurabh Gupta Seunghoon Hong Laszlo Jeni
Tanmay Gupta Wei Hong Hueihan Jhuang
Abner Guzman Rivera Anthony Hoogs Dinghuang Ji
Timo Hackel Radu Horaud Hui Ji
Sunil Hadap Yedid Hoshen Qiang Ji
Christian Haene Omid Hosseini Jafari Fan Jia
Ralf Haeusler Kuang-Jui Hsu Kui Jia
Levente Hajder Winston Hsu Xu Jia
David Hall Yinlin Hu Huaizu Jiang
Peter Hall Zhe Hu Jiayan Jiang
Stefan Haller Gang Hua Nianjuan Jiang
Ghassan Hamarneh Chen Huang Tingting Jiang
Fred Hamprecht De-An Huang Xiaoyi Jiang
Onur Hamsici Dong Huang Yu-Gang Jiang
Bohyung Han Gary Huang Long Jin
Junwei Han Heng Huang Suo Jinli
Xufeng Han Jia-Bin Huang Justin Johnson
Yahong Han Qixing Huang Nebojsa Jojic
Ankur Handa Rui Huang Michael Jones
Albert Haque Sheng Huang Hanbyul Joo
Tatsuya Harada Weilin Huang Jungseock Joo
Mehrtash Harandi Xiaolei Huang Ajjen Joshi
Bharath Hariharan Xinyu Huang Amin Jourabloo
Mahmudul Hasan Zhiwu Huang Frederic Jurie
Tal Hassner Tak-Wai Hui Achuta Kadambi
Kenji Hata Wei-Chih Hung Samuel Kadoury
Soren Hauberg Junhwa Hur Ioannis Kakadiaris
Michal Havlena Mohamed Hussein Zdenek Kalal
Zeeshan Hayder Wonjun Hwang Yannis Kalantidis
Junfeng He Anders Hyden Sinan Kalkan
Lei He Satoshi Ikehata Vicky Kalogeiton
Varsha Hedau Nazlı Ikizler-Cinbis Sunkavalli Kalyan
Felix Heide Viorela Ila J.-K. Kamarainen
Organization XIX
Martin Kampel Dimitrios Kosmopoulos Victor Lempitsky

Kenichi Kanatani Satwik Kottur Spyridon Leonardos
Angjoo Kanazawa Balazs Kovacs Marius Leordeanu
Melih Kandemir Adarsh Kowdle Matt Leotta
Sing Bing Kang Mike Krainin Thomas Leung
Zhuoliang Kang Gregory Kramida Stefan Leutenegger
Mohan Kankanhalli Ranjay Krishna Gil Levi
Juho Kannala Ravi Krishnan Aviad Levis
Abhishek Kar Matej Kristan Jose Lezama
Amlan Kar Pavel Krsek Ang Li
Svebor Karaman Volker Krueger Dingzeyu Li
Leonid Karlinsky Alexander Krull Dong Li
Zoltan Kato Hilde Kuehne Haoxiang Li
Parneet Kaur Andreas Kuhn Hongdong Li
Hiroshi Kawasaki Arjan Kuijper Hongsheng Li
Misha Kazhdan Zuzana Kukelova Hongyang Li
Margret Keuper Kuldeep Kulkarni Jianguo Li
Sameh Khamis Shiro Kumano Kai Li
Naeemullah Khan Avinash Kumar Ruiyu Li
Salman Khan Vijay Kumar Wei Li
Hadi Kiapour Abhijit Kundu Wen Li
Joe Kileel Sebastian Kurtek Xi Li
Chanho Kim Junseok Kwon Xiaoxiao Li
Gunhee Kim Jan Kybic Xin Li
Hansung Kim Alexander Ladikos Xirong Li
Junmo Kim Shang-Hong Lai Xuelong Li
Junsik Kim Wei-Sheng Lai Xueting Li
Kihwan Kim Jean-Francois Lalonde Yeqing Li
Minyoung Kim John Lambert Yijun Li
Tae Hyun Kim Zhenzhong Lan Yin Li
Tae-Kyun Kim Charis Lanaras Yingwei Li
Akisato Kimura Oswald Lanz Yining Li
Zsolt Kira Dong Lao Yongjie Li
Alexander Kirillov Longin Jan Latecki Yu-Feng Li
Kris Kitani Justin Lazarow Zechao Li
Maria Klodt Huu Le Zhengqi Li
Patrick Knöbelreiter Chen-Yu Lee Zhenyang Li
Jan Knopp Gim Hee Lee Zhizhong Li
Reinhard Koch Honglak Lee Xiaodan Liang
Alexander Kolesnikov Hsin-Ying Lee Renjie Liao
Chen Kong Joon-Young Lee Zicheng Liao
Naejin Kong Seungyong Lee Bee Lim
Shu Kong Stefan Lee Jongwoo Lim
Piotr Koniusz Yong Jae Lee Joseph Lim
Simon Korman Zhen Lei Ser-Nam Lim
Andreas Koschan Ido Leichter Chen-Hsuan Lin
XX Organization
Shih-Yao Lin Simon Lucey Christopher Mei

Tsung-Yi Lin Jian-Hao Luo Heydi Mendez-Vazquez
Weiyao Lin Jiebo Luo Deyu Meng
Yen-Yu Lin Pablo Márquez-Neila Thomas Mensink
Haibin Ling Matthias Müller Bjoern Menze
Or Litany Chao Ma Domingo Mery
Roee Litman Chih-Yao Ma Qiguang Miao
Anan Liu Lin Ma Tomer Michaeli
Changsong Liu Shugao Ma Antoine Miech
Chen Liu Wei-Chiu Ma Ondrej Miksik
Ding Liu Zhanyu Ma Anton Milan
Dong Liu Oisin Mac Aodha Gregor Miller
Feng Liu Will Maddern Cai Minjie
Guangcan Liu Ludovic Magerand Majid Mirmehdi
Luoqi Liu Marcus Magnor Ishan Misra
Miaomiao Liu Vijay Mahadevan Niloy Mitra
Nian Liu Mohammad Mahoor Anurag Mittal
Risheng Liu Michael Maire Nirbhay Modhe
Shu Liu Subhransu Maji Davide Modolo
Shuaicheng Liu Ameesh Makadia Pritish Mohapatra
Sifei Liu Atsuto Maki Pascal Monasse
Tyng-Luh Liu Yasushi Makihara Mathew Monfort
Wanquan Liu Mateusz Malinowski Taesup Moon
Weiwei Liu Tomasz Malisiewicz Sandino Morales
Xialei Liu Arun Mallya Vlad Morariu
Xiaoming Liu Roberto Manduchi Philippos Mordohai
Yebin Liu Junhua Mao Francesc Moreno
Yiming Liu Dmitrii Marin Henrique Morimitsu
Ziwei Liu Joe Marino Yael Moses
Zongyi Liu Kenneth Marino Ben-Ezra Moshe
Liliana Lo Presti Elisabeta Marinoiu Roozbeh Mottaghi
Edgar Lobaton Ricardo Martin Yadong Mu
Chengjiang Long Aleix Martinez Lopamudra Mukherjee
Mingsheng Long Julieta Martinez Mario Munich
Roberto Lopez-Sastre Aaron Maschinot Ana Murillo
Amy Loufti Jonathan Masci Damien Muselet
Brian Lovell Bogdan Matei Armin Mustafa
Canyi Lu Diana Mateus Siva Karthik Mustikovela
Cewu Lu Stefan Mathe Moin Nabi
Feng Lu Kevin Matzen Sobhan Naderi
Huchuan Lu Bruce Maxwell Hajime Nagahara
Jiajun Lu Steve Maybank Varun Nagaraja
Jiasen Lu Walterio Mayol-Cuevas Tushar Nagarajan
Jiwen Lu Mason McGill Arsha Nagrani
Yang Lu Stephen Mckenna Nikhil Naik
Yujuan Lu Roey Mechrez Atsushi Nakazawa
Organization XXI
P. J. Narayanan Hyun Soo Park Victor Prisacariu

Charlie Nash In Kyu Park Jan Prokaj
Lakshmanan Nataraj Jaesik Park Nicolas Pugeault
Fabian Nater Omkar Parkhi Luis Puig
Lukáš Neumann Alvaro Parra Bustos Ali Punjani
Natalia Neverova C. Alejandro Parraga Senthil Purushwalkam
Alejandro Newell Vishal Patel Guido Pusiol
Phuc Nguyen Deepak Pathak Guo-Jun Qi
Xiaohan Nie Ioannis Patras Xiaojuan Qi
David Nilsson Viorica Patraucean Hongwei Qin
Ko Nishino Genevieve Patterson Shi Qiu
Zhenxing Niu Kim Pedersen Faisal Qureshi
Shohei Nobuhara Robert Peharz Matthias Rüther
Klas Nordberg Selen Pehlivan Petia Radeva
Mohammed Norouzi Xi Peng Umer Rafi
David Novotny Bojan Pepik Rahul Raguram
Ifeoma Nwogu Talita Perciano Swaminathan Rahul
Matthew O’Toole Federico Pernici Varun Ramakrishna
Guillaume Obozinski Adrian Peter Kandan Ramakrishnan
Jean-Marc Odobez Stavros Petridis Ravi Ramamoorthi
Eyal Ofek Vladimir Petrovic Vignesh Ramanathan
Ferda Ofli Henning Petzka Vasili Ramanishka
Tae-Hyun Oh Tomas Pfister R. Ramasamy Selvaraju
Iason Oikonomidis Trung Pham Rene Ranftl
Takeshi Oishi Justus Piater Carolina Raposo
Takahiro Okabe Massimo Piccardi Nikhil Rasiwasia
Takayuki Okatani Sudeep Pillai Nalini Ratha
Vlad Olaru Pedro Pinheiro Sai Ravela
Michael Opitz Lerrel Pinto Avinash Ravichandran
Jose Oramas Bernardo Pires Ramin Raziperchikolaei
Vicente Ordonez Aleksis Pirinen Sylvestre-Alvise Rebuffi
Ivan Oseledets Fiora Pirri Adria Recasens
Aljosa Osep Leonid Pischulin Joe Redmon
Magnus Oskarsson Tobias Ploetz Timo Rehfeld
Martin R. Oswald Bryan Plummer Michal Reinstein
Wanli Ouyang Yair Poleg Konstantinos Rematas
Andrew Owens Jean Ponce Haibing Ren
Mustafa Özuysal Gerard Pons-Moll Shaoqing Ren
Jinshan Pan Jordi Pont-Tuset Wenqi Ren
Xingang Pan Alin Popa Zhile Ren
Rameswar Panda Fatih Porikli Hamid Rezatofighi
Sharath Pankanti Horst Possegger Nicholas Rhinehart
Julien Pansiot Viraj Prabhu Helge Rhodin
Nicolas Papadakis Andrea Prati Elisa Ricci
George Papandreou Maria Priisalu Eitan Richardson
N. Papanikolopoulos Véronique Prinet Stephan Richter
XXII Organization
Gernot Riegler Torsten Sattler Tianmin Shu

Hayko Riemenschneider Bogdan Savchynskyy Zhixin Shu
Tammy Riklin Raviv Johannes Schönberger Kaleem Siddiqi
Ergys Ristani Hanno Scharr Gunnar Sigurdsson
Tobias Ritschel Walter Scheirer Nathan Silberman
Mariano Rivera Bernt Schiele Tomas Simon
Samuel Rivera Frank Schmidt Abhishek Singh
Antonio Robles-Kelly Tanner Schmidt Gautam Singh
Ignacio Rocco Dirk Schnieders Maneesh Singh
Jason Rock Samuel Schulter Praveer Singh
Emanuele Rodola William Schwartz Richa Singh
Mikel Rodriguez Alexander Schwing Saurabh Singh
Gregory Rogez Ozan Sener Sudipta Sinha
Marcus Rohrbach Soumyadip Sengupta Vladimir Smutny
Gemma Roig Laura Sevilla-Lara Noah Snavely
Javier Romero Mubarak Shah Cees Snoek
Olaf Ronneberger Shishir Shah Kihyuk Sohn
Amir Rosenfeld Fahad Shahbaz Khan Eric Sommerlade
Bodo Rosenhahn Amir Shahroudy Sanghyun Son
Guy Rosman Jing Shao Bi Song
Arun Ross Xiaowei Shao Shiyu Song
Samuel Rota Bulò Roman Shapovalov Shuran Song
Peter Roth Nataliya Shapovalova Xuan Song
Constantin Rothkopf Ali Sharif Razavian Yale Song
Sebastien Roy Gaurav Sharma Yang Song
Amit Roy-Chowdhury Mohit Sharma Yibing Song
Ognjen Rudovic Pramod Sharma Lorenzo Sorgi
Adria Ruiz Viktoriia Sharmanska Humberto Sossa
Javier Ruiz-del-Solar Eli Shechtman Pratul Srinivasan
Christian Rupprecht Mark Sheinin Michael Stark
Olga Russakovsky Evan Shelhamer Bjorn Stenger
Chris Russell Chunhua Shen Rainer Stiefelhagen
Alexandre Sablayrolles Li Shen Joerg Stueckler
Fereshteh Sadeghi Wei Shen Jan Stuehmer
Ryusuke Sagawa Xiaohui Shen Hang Su
Hideo Saito Xiaoyong Shen Hao Su
Elham Sakhaee Ziyi Shen Shuochen Su
Albert Ali Salah Lu Sheng R. Subramanian
Conrad Sanderson Baoguang Shi Yusuke Sugano
Koppal Sanjeev Boxin Shi Akihiro Sugimoto
Aswin Sankaranarayanan Kevin Shih Baochen Sun
Elham Saraee Hyunjung Shim Chen Sun
Jason Saragih Ilan Shimshoni Jian Sun
Sudeep Sarkar Young Min Shin Jin Sun
Imari Sato Koichi Shinoda Lin Sun
Shin’ichi Satoh Matthew Shreve Min Sun
Organization XXIII
Qing Sun Chetan Tonde Matthias Vestner

Zhaohui Sun Xin Tong Minh Vo
David Suter Akihiko Torii Christoph Vogel
Eran Swears Andrea Torsello Michele Volpi
Raza Syed Hussain Florian Trammer Carl Vondrick
T. Syeda-Mahmood Du Tran Sven Wachsmuth
Christian Szegedy Quoc-Huy Tran Toshikazu Wada
Duy-Nguyen Ta Rudolph Triebel Michael Waechter
Tolga Taşdizen Alejandro Troccoli Catherine Wah
Hemant Tagare Leonardo Trujillo Jacob Walker
Yuichi Taguchi Tomasz Trzcinski Jun Wan
Ying Tai Sam Tsai Boyu Wang
Yu-Wing Tai Yi-Hsuan Tsai Chen Wang
Jun Takamatsu Hung-Yu Tseng Chunyu Wang
Hugues Talbot Vagia Tsiminaki De Wang
Toru Tamak Aggeliki Tsoli Fang Wang
Robert Tamburo Wei-Chih Tu Hongxing Wang
Chaowei Tan Shubham Tulsiani Hua Wang
Meng Tang Fred Tung Jiang Wang
Peng Tang Tony Tung Jingdong Wang
Siyu Tang Matt Turek Jinglu Wang
Wei Tang Oncel Tuzel Jue Wang
Junli Tao Georgios Tzimiropoulos Le Wang
Ran Tao Ilkay Ulusoy Lei Wang
Xin Tao Osman Ulusoy Lezi Wang
Makarand Tapaswi Dmitry Ulyanov Liang Wang
Jean-Philippe Tarel Paul Upchurch Lichao Wang
Maxim Tatarchenko Ben Usman Lijun Wang
Bugra Tekin Evgeniya Ustinova Limin Wang
Demetri Terzopoulos Himanshu Vajaria Liwei Wang
Christian Theobalt Alexander Vakhitov Naiyan Wang
Diego Thomas Jack Valmadre Oliver Wang
Rajat Thomas Ernest Valveny Qi Wang
Qi Tian Jan van Gemert Ruiping Wang
Xinmei Tian Grant Van Horn Shenlong Wang
YingLi Tian Jagannadan Varadarajan Shu Wang
Yonghong Tian Gul Varol Song Wang
Yonglong Tian Sebastiano Vascon Tao Wang
Joseph Tighe Francisco Vasconcelos Xiaofang Wang
Radu Timofte Mayank Vatsa Xiaolong Wang
Massimo Tistarelli Javier Vazquez-Corral Xinchao Wang
Sinisa Todorovic Ramakrishna Vedantam Xinggang Wang
Pavel Tokmakov Ashok Veeraraghavan Xintao Wang
Giorgos Tolias Andreas Veit Yang Wang
Federico Tombari Raviteja Vemulapalli Yu-Chiang Frank Wang
Tatiana Tommasi Jonathan Ventura Yu-Xiong Wang
XXIV Organization
Zhaowen Wang Jin Xie Michael Ying Yang

Zhe Wang Lingxi Xie Ming Yang
Anne Wannenwetsch Pengtao Xie Ruiduo Yang
Simon Warfield Saining Xie Ruigang Yang
Scott Wehrwein Wenxuan Xie Shuo Yang
Donglai Wei Yuchen Xie Wei Yang
Ping Wei Bo Xin Xiaodong Yang
Shih-En Wei Junliang Xing Yanchao Yang
Xiu-Shen Wei Peng Xingchao Yi Yang
Yichen Wei Bo Xiong Angela Yao
Xie Weidi Fei Xiong Bangpeng Yao
Philippe Weinzaepfel Xuehan Xiong Cong Yao
Longyin Wen Yuanjun Xiong Jian Yao
Eric Wengrowski Chenliang Xu Ting Yao
Tomas Werner Danfei Xu Julian Yarkony
Michael Wilber Huijuan Xu Mark Yatskar
Rick Wildes Jia Xu Jinwei Ye
Olivia Wiles Weipeng Xu Mao Ye
Kyle Wilson Xiangyu Xu Mei-Chen Yeh
David Wipf Yan Xu Raymond Yeh
Kwan-Yee Wong Yuanlu Xu Serena Yeung
Daniel Worrall Jia Xue Kwang Moo Yi
John Wright Tianfan Xue Shuai Yi
Baoyuan Wu Erdem Yörük Alper Yılmaz
Chao-Yuan Wu Abhay Yadav Lijun Yin
Jiajun Wu Deshraj Yadav Xi Yin
Jianxin Wu Payman Yadollahpour Zhaozheng Yin
Tianfu Wu Yasushi Yagi Xianghua Ying
Xiaodong Wu Toshihiko Yamasaki Ryo Yonetani
Xiaohe Wu Fei Yan Donghyun Yoo
Xinxiao Wu Hang Yan Ju Hong Yoon
Yang Wu Junchi Yan Kuk-Jin Yoon
Yi Wu Junjie Yan Chong You
Ying Wu Sijie Yan Shaodi You
Yuxin Wu Keiji Yanai Aron Yu
Zheng Wu Bin Yang Fisher Yu
Stefanie Wuhrer Chih-Yuan Yang Gang Yu
Yin Xia Dong Yang Jingyi Yu
Tao Xiang Herb Yang Ke Yu
Yu Xiang Jianchao Yang Licheng Yu
Lei Xiao Jianwei Yang Pei Yu
Tong Xiao Jiaolong Yang Qian Yu
Yang Xiao Jie Yang Rong Yu
Cihang Xie Jimei Yang Shoou-I Yu
Dan Xie Jufeng Yang Stella Yu
Jianwen Xie Linjie Yang Xiang Yu
Organization XXV
Yang Yu Quanshi Zhang Guang-Tong Zhou

Zhiding Yu Richard Zhang Huiyu Zhou
Ganzhao Yuan Runze Zhang Jiahuan Zhou
Jing Yuan Shanshan Zhang S. Kevin Zhou
Junsong Yuan Shiliang Zhang Tinghui Zhou
Lu Yuan Shu Zhang Wengang Zhou
Stefanos Zafeiriou Ting Zhang Xiaowei Zhou
Sergey Zagoruyko Xiangyu Zhang Xingyi Zhou
Amir Zamir Xiaofan Zhang Yin Zhou
K. Zampogiannis Xu Zhang Zihan Zhou
Andrei Zanfir Yimin Zhang Fan Zhu
Mihai Zanfir Yinda Zhang Guangming Zhu
Pablo Zegers Yongqiang Zhang Ji Zhu
Eyasu Zemene Yuting Zhang Jiejie Zhu
Andy Zeng Zhanpeng Zhang Jun-Yan Zhu
Xingyu Zeng Ziyu Zhang Shizhan Zhu
Yun Zeng Bin Zhao Siyu Zhu
De-Chuan Zhan Chen Zhao Xiangxin Zhu
Cheng Zhang Hang Zhao Xiatian Zhu
Dong Zhang Hengshuang Zhao Yan Zhu
Guofeng Zhang Qijun Zhao Yingying Zhu
Han Zhang Rui Zhao Yixin Zhu
Hang Zhang Yue Zhao Yuke Zhu
Hanwang Zhang Enliang Zheng Zhenyao Zhu
Jian Zhang Liang Zheng Liansheng Zhuang
Jianguo Zhang Stephan Zheng Zeeshan Zia
Jianming Zhang Wei-Shi Zheng Karel Zimmermann
Jiawei Zhang Wenming Zheng Daniel Zoran
Junping Zhang Yin Zheng Danping Zou
Lei Zhang Yinqiang Zheng Qi Zou
Linguang Zhang Yuanjie Zheng Silvia Zuffi
Ning Zhang Guangyu Zhong Wangmeng Zuo
Qing Zhang Bolei Zhou Xinxin Zuo
Contents – Part I
Learning for Vision
Convolutional Networks with Adaptive Inference Graphs . . . . . . . . . . . . . . . 3

Andreas Veit and Serge Belongie
Progressive Neural Architecture Search . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Chenxi Liu, Barret Zoph, Maxim Neumann, Jonathon Shlens, Wei Hua,
Li-Jia Li, Li Fei-Fei, Alan Yuille, Jonathan Huang, and Kevin Murphy
Diverse Image-to-Image Translation via Disentangled Representations . . . . . . 36

Hsin-Ying Lee, Hung-Yu Tseng, Jia-Bin Huang, Maneesh Singh,
and Ming-Hsuan Yang
Lifting Layers: Analysis and Applications . . . . . . . . . . . . . . . . . . . . . . . . . 53

Peter Ochs, Tim Meinhardt, Laura Leal-Taixe, and Michael Moeller
Learning with Biased Complementary Labels . . . . . . . . . . . . . . . . . . . . . . . 69

Xiyu Yu, Tongliang Liu, Mingming Gong, and Dacheng Tao
Poster Session
Semi-convolutional Operators for Instance Segmentation . . . . . . . . . . . . . . . 89

David Novotny, Samuel Albanie, Diane Larlus, and Andrea Vedaldi
Skeleton-Based Action Recognition with Spatial Reasoning and Temporal

Stack Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Chenyang Si, Ya Jing, Wei Wang, Liang Wang, and Tieniu Tan
Fictitious GAN: Training GANs with Historical Models . . . . . . . . . . . . . . . 122

Hao Ge, Yin Xia, Xu Chen, Randall Berry, and Ying Wu
Bi-box Regression for Pedestrian Detection and Occlusion Estimation . . . . . . 138

Chunluan Zhou and Junsong Yuan
C-WSL: Count-Guided Weakly Supervised Localization . . . . . . . . . . . . . . . 155

Mingfei Gao, Ang Li, Ruichi Yu, Vlad I. Morariu, and Larry S. Davis
Attributes as Operators: Factorizing Unseen

Attribute-Object Compositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Tushar Nagarajan and Kristen Grauman
Product Quantization Network for Fast Image Retrieval . . . . . . . . . . . . . . . . 191

Tan Yu, Junsong Yuan, Chen Fang, and Hailin Jin
XXVIII Contents – Part I
Cross-Modal Hamming Hashing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

Yue Cao, Bin Liu, Mingsheng Long, and Jianmin Wang
Deep Video Quality Assessor: From Spatio-Temporal Visual Sensitivity

to a Convolutional Neural Aggregation Network . . . . . . . . . . . . . . . . . . . . . 224
Woojae Kim, Jongyoo Kim, Sewoong Ahn, Jinwoo Kim,
and Sanghoon Lee
Semi-dense 3D Reconstruction with a Stereo Event Camera . . . . . . . . . . . . . 242

Yi Zhou, Guillermo Gallego, Henri Rebecq, Laurent Kneip,
Hongdong Li, and Davide Scaramuzza
Self-Calibrating Isometric Non-Rigid Structure-from-Motion. . . . . . . . . . . . . 259

Shaifali Parashar, Adrien Bartoli, and Daniel Pizarro
Semi-supervised Deep Learning with Memory . . . . . . . . . . . . . . . . . . . . . . 275

Yanbei Chen, Xiatian Zhu, and Shaogang Gong
Deep Fundamental Matrix Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

René Ranftl and Vladlen Koltun
TrackingNet: A Large-Scale Dataset and Benchmark for Object Tracking

in the Wild. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
Matthias Müller, Adel Bibi, Silvio Giancola, Salman Alsubaihi,
and Bernard Ghanem
StarMap for Category-Agnostic Keypoint and Viewpoint Estimation . . . . . . . 328

Xingyi Zhou, Arjun Karpur, Linjie Luo, and Qixing Huang
Factorizable Net: An Efficient Subgraph-Based Framework for Scene

Graph Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
Yikang Li, Wanli Ouyang, Bolei Zhou, Jianping Shi, Chao Zhang,
and Xiaogang Wang
Multi-fiber Networks for Video Recognition. . . . . . . . . . . . . . . . . . . . . . . . 364

Yunpeng Chen, Yannis Kalantidis, Jianshu Li, Shuicheng Yan,
and Jiashi Feng
Tackling 3D ToF Artifacts Through Learning and the FLAT Dataset. . . . . . . 381
Qi Guo, Iuri Frosio, Orazio Gallo, Todd Zickler, and Jan Kautz
Zero-Shot Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397

Ankan Bansal, Karan Sikka, Gaurav Sharma, Rama Chellappa,
and Ajay Divakaran
Contents – Part I XXIX
A Modulation Module for Multi-task Learning with Applications

in Image Retrieval. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Xiangyun Zhao, Haoxiang Li, Xiaohui Shen, Xiaodan Liang,
and Ying Wu
Fast and Accurate Intrinsic Symmetry Detection . . . . . . . . . . . . . . . . . . . . . 433

Rajendra Nagar and Shanmuganathan Raman
Objects that Sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451

Relja Arandjelović and Andrew Zisserman
Deblurring Natural Image Using Super-Gaussian Fields . . . . . . . . . . . . . . . . 467

Yuhang Liu, Wenyong Dong, Dong Gong, Lei Zhang, and Qinfeng Shi
Question-Guided Hybrid Convolution for Visual Question Answering . . . . . . 485

Peng Gao, Hongsheng Li, Shuang Li, Pan Lu, Yikang Li,
Steven C. H. Hoi, and Xiaogang Wang
Geometric Constrained Joint Lane Segmentation and Lane

Boundary Detection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
Jie Zhang, Yi Xu, Bingbing Ni, and Zhenyu Duan
Unpaired Image Captioning by Language Pivoting . . . . . . . . . . . . . . . . . . . 519

Jiuxiang Gu, Shafiq Joty, Jianfei Cai, and Gang Wang
Efficient Uncertainty Estimation for Semantic Segmentation in Videos . . . . . 536

Po-Yu Huang, Wan-Ting Hsu, Chun-Yueh Chiu, Ting-Fan Wu,
and Min Sun
Person Search by Multi-Scale Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . 553

Xu Lan, Xiatian Zhu, and Shaogang Gong
A Hybrid Model for Identity Obfuscation by Face Replacement . . . . . . . . . . 570

Qianru Sun, Ayush Tewari, Weipeng Xu, Mario Fritz,
Christian Theobalt, and Bernt Schiele
The Sound of Pixels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587

Hang Zhao, Chuang Gan, Andrew Rouditchenko, Carl Vondrick,
Josh McDermott, and Antonio Torralba
Adaptive Affinity Fields for Semantic Segmentation . . . . . . . . . . . . . . . . . . 605

Tsung-Wei Ke, Jyh-Jing Hwang, Ziwei Liu, and Stella X. Yu
ReenactGAN: Learning to Reenact Faces via Boundary Transfer. . . . . . . . . . 622

Wayne Wu, Yunxuan Zhang, Cheng Li, Chen Qian,
and Chen Change Loy
Learning to Anonymize Faces for Privacy Preserving Action Detection . . . . . 639

Zhongzheng Ren, Yong Jae Lee, and Michael S. Ryoo
XXX Contents – Part I
Joint Person Segmentation and Identification in Synchronized

First- and Third-Person Videos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 656
Mingze Xu, Chenyou Fan, Yuchen Wang, Michael S. Ryoo,
and David J. Crandall
Neural Graph Matching Networks for Fewshot 3D Action Recognition . . . . . 673

Michelle Guo, Edward Chou, De-An Huang, Shuran Song,
Serena Yeung, and Li Fei-Fei
Graph R-CNN for Scene Graph Generation . . . . . . . . . . . . . . . . . . . . . . . . 690

Jianwei Yang, Jiasen Lu, Stefan Lee, Dhruv Batra, and Devi Parikh
Deep Cross-Modal Projection Learning for Image-Text Matching . . . . . . . . . 707

Ying Zhang and Huchuan Lu
ShapeStacks: Learning Vision-Based Physical Intuition for Generalised

Object Stacking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724
Oliver Groth, Fabian B. Fuchs, Ingmar Posner, and Andrea Vedaldi
Inner Space Preserving Generative Pose Machine . . . . . . . . . . . . . . . . . . . . 740

Shuangjun Liu and Sarah Ostadabbas
Attention-Based Ensemble for Deep Metric Learning. . . . . . . . . . . . . . . . . . 760

Wonsik Kim, Bhavya Goyal, Kunal Chawla, Jungmin Lee,
and Keunjoo Kwon
Learning Compression from Limited Unlabeled Data . . . . . . . . . . . . . . . . . . 778

Xiangyu He and Jian Cheng
Discriminative Region Proposal Adversarial Networks for High-Quality

Image-to-Image Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 796
Chao Wang, Haiyong Zheng, Zhibin Yu, Ziqiang Zheng, Zhaorui Gu,
and Bing Zheng
Unsupervised Video Object Segmentation Using Motion Saliency-Guided

Spatio-Temporal Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 813
Yuan-Ting Hu, Jia-Bin Huang, and Alexander G. Schwing
Temporal Relational Reasoning in Videos . . . . . . . . . . . . . . . . . . . . . . . . . 831

Bolei Zhou, Alex Andonian, Aude Oliva, and Antonio Torralba
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 847

Learning for Vision
Another random document with
no related content on Scribd:
order of Asteroidea to which Asterias belongs, the adambulacrals
themselves do not project much, but in all other cases they form
prominent mouth-angles, so that the opening of the mouth becomes
star-shaped (Fig. 211, p. 483).
Except in the case of the ambulacral and adambulacral plates little

regular arrangement is to be detected in the ossicles of the skeleton
which, as has already been mentioned, form a mesh-work. If,
however, the arm be cut open and viewed from the inside it will be
seen that the edge is strengthened above and below by very thick,
powerful, rod-like plates. These are called the "supero-marginal" and
"infero-marginal" ossicles; they are not visible from the outside, since
they are covered by a thick layer of the body-wall containing other
smaller plates (Fig. 190, marg). In many genera, however, they are
exposed, and form a conspicuous edging to the arm above and
below. In many genera, also, there are three conspicuous series of
plates on the back of each arm, viz. a median row, called "carinals"
(car., Fig. 191), and two lateral rows, termed "dorso-laterals" (d.lat.,
Fig. 191). These three rows, with the two rows of marginals, one of
ambulacrals, and one of adambulacrals on each side (11 rows in all),
constitute the primitive skeleton of the arm, and appear first in
development.
The structure of all these elements of the skeleton is the same. They
may be described as scaffoldings of carbonate of lime,
interpenetrated by a mesh-work of cells fused with one another, by
which the carbonate of lime has been deposited. The matrix in which
the ossicles lie is a jelly-like substance traversed by a few bands of
fibres which connect the various rods with one another. This jelly is
almost fluid in the fresh state, but when heated forms a hard
compound, possibly allied to mucin, which will turn the edge of a
razor.
When the covering of the back is dissected off the coelom is

opened. This is a spacious cavity which apparently surrounds the
alimentary canal and extends into the arms. It has, however, its own
proper wall, which is called the "peritoneum," both on the outer side,
where it abuts on the skin, and on the inner side, where it comes in
contact with the wall of the alimentary canal. The outer wall is called
the "somatic peritoneum," and it is possible to dissect off the rest of
the body-wall and leave it intact; the inner wall, from its close
association with the alimentary canal, is termed the "splanchnic
peritoneum." This wall can only be distinguished in microscopic
sections from that of the alimentary canal, to which it is closely
applied.
The coelom is filled with a fluid, which is practically sea water with a
little albuminous matter in solution. Through the thin walls of the
papulae oxygen passes into this fluid, whence it easily reaches the
inner organs, since they are all in contact with some part of the
coelomic wall. Similarly CO2 is absorbed by the coelomic fluid from
all parts of the body, and diffuses through the papulae to the
surrounding water.
The Starfish possesses no definite kidney for getting rid of

nitrogenous waste. In most of the higher animals with a well-
developed coelom it has been proved that the kidney is simply a
specialised portion of the coelom, and in many cases some parts of
the coelomic wall still retain their excretory functions, which
apparently the whole originally possessed. In the Starfish and in
Echinodermata generally this primitive state of affairs is still retained.
From the cells forming the coelomic wall, cells are budded off into
the fluid, where they swim about. These cells from their movements
are called amoebocytes. If a substance such as indigo-carmine,
which when introduced into the tissues of the higher animals is
eliminated by the kidney, is injected into the Starfish, it is found soon
after to be vigorously absorbed by the amoebocytes. These later
accumulate in the dermal branchiae, through the thin walls of which
they make their way[442] to the outside, where they degenerate.
The coelom is indented by five folds, which project inwards from the
interradii. These folds are called the "interradial septa"; they are
stiffened by a calcareous deposit, which is not, however, sufficiently
dense to constitute a plate. In one of the septa the axial sinus and
stone-canal (see below) are embedded. These septa are to be
regarded as areas of lateral adhesion between the arms.
Fig. 188.—View of upper half of a specimen of Asterias rubens, which has been
split horizontally into two halves. ax.c, Axial sinus; g.d, genital duct; oe, cut
end of the oesophagus, the narrow neck of the stomach; py, pyloric sac;
py.c, pyloric caeca; r, rectum; r.c, rectal caeca; sept, interradial septum; st.c,
stomach lobe.
The alimentary canal consists of several distinct portions. The

mouth leads by a narrow neck called the "oesophagus" into a
voluminous baggy sac termed the "stomach," which is produced into
ten short pouches, two projecting into each arm. The stomach leads
in turn by a wide opening into a pentagonal flattened sac, the "pyloric
sac," which lies above it. Each angle of the pyloric sac is prolonged
into a tube—the so-called "pyloric duct"—running out into the arm,
where it immediately bifurcates into two forks, each beset by a large
number of small pouches and attached to the dorsal wall of the
coelom by suspensory bands of membrane called mesenteries.
These ten forks are called "pyloric caeca"; they are of a deep green
colour owing to the pigment in their wall. Beyond the pyloric sac the
alimentary canal is continued as the slender "rectum" to the anus.
The rectum gives off two small branched pouches of a brown colour
called "rectal caeca." This comparatively complicated form of
alimentary canal is related to the nature of the food of the animal and
the method it employs to capture its prey.
Fig. 189.—View of a Starfish (Echinaster) devouring a Mussel. 1. The

madreporite.
The favourite food[443] of Asterias consists of the common bivalves

of the coast, notably of the Mussel (Mytilus edulis). There is,
however, no animal which it will not attack if it is fortunate enough to
be able to catch it. The Starfish seizes its prey by the tube-feet, and
places it directly under its mouth, folding its arms down over it in
umbrella fashion. The muscles which run around the arms and disc
in the body-wall contract, and the pressure thus brought to bear on
the incompressible fluid contained in the coelom, forces out the thin
membranous peristome and partially turns the stomach inside out.
The everted edge of the stomach is wrapped round the prey.
Soon the bivalve is forced to relax its muscles and allow the valves
to gape. The edge of the stomach is then inserted between the
valves and applied directly to the soft parts of the prey which is thus
completely digested. When the Starfish moves away nothing but the
cleaned shell is left behind. If the bivalve is small it may be
completely taken into the stomach, and the empty shell later rejected
through the mouth.
It was for a long time a puzzle in what way the bivalve was forced to
open. Schiemenz[444] has, however, shown that when the Starfish
folds itself in umbrella-like form over the prey it holds on to the
substratum by means of the tube-feet of the distal portions of the
arms, whilst, by means of the tube-feet belonging to the central
portions, it drags apart the valves by main force. He has shown
experimentally: (1) that whilst a bivalve may be able to resist a
sudden pull of 4000 grammes it will yield to a pull of 900 grammes
long continued; (2) that a Starfish can exert a pull of 1350 grammes;
(3) that a Starfish is unable to open a bivalve unless it be allowed to
raise itself into a hump, so that the pull of the central tube-feet is at
right angles to the prey. A Starfish confined between two glass plates
walked about all day carrying with it a bivalve which it was unable to
open.
The lining of the stomach is found to consist very largely of mucus-

forming cells, which are swollen with large drops of mucus or some
similar substance. It used to be supposed that this substance had
some poisonous action on the prey and paralysed it, but the
researches of Schiemenz show that this is incorrect. If when an
Asterias is devouring a bivalve another be offered to it, it will open it,
but will not digest it, and the victim shows no sign of injury but soon
recovers. The cells forming the walls of the pyloric sac and its
appendages are tall narrow cylindrical cells crowded with granules
which appear to be of the nature of digestive ferment. This
substance flows into the stomach and digests the captured prey.
A very small amount of matter passes into the rectum and escapes
by the anus, as the digestive powers of the Starfish are very
complete. The rectal caeca are lined by cells which secrete from the
coelomic fluid a brown material, in all probability an excretion, which
is got rid of by the anus.
When the meal is finished the stomach is restored to its former place
by the action of five pairs of retractor muscles, one pair of which
originates from the upper surface of the ambulacral ossicles in each
arm and extends to the wall of the stomach, where they are inserted
(Fig. 190, ret).
The tube-feet, which are at once the locomotor and the principal
sensory organs of the Starfish, are appendages of that peculiar
system of tubes known as the water-vascular system, which is
derived from a part of the coelom cut off from the rest during the
development of the animal. This system, as already mentioned,
consists of (1) a narrow "ring-canal," encircling the mouth and lying
on the inner surface of the membranous peristome; (2) a radial canal
leaving the ring-canal and running along the under surface of each
arm just above the ambulacral groove; (3) a vertical stone-canal
running from the madreporite downwards to open into the ring-canal
in the interspace between two arms. The madreporite is covered
externally by grooves lined with long cilia, and is pierced with narrow
canals of excessively fine calibre, the walls of which are also lined by
powerful cilia. Most of these narrow canals open below into a main
collecting canal, the stone-canal, but some open into a division of the
coelom termed the axial sinus, with which also the stone-canal
communicates by a lateral opening. The cavity of the stone-canal is
reduced by the outgrowth from its walls of a peculiar Y-shaped
projection, the ends being rolled on themselves in a complicated way
(Fig. 190, B). The walls of the canal consist of a layer of very long
narrow cells, which carry powerful flagella, and outside this of a crust
of calcareous deposit, which gives rigidity to the walls and has
suggested the name stone-canal.
The tube-feet are covered externally by ectoderm, inside which is a

tube in connexion with the radial water-vascular canal. This latter is
lined by flattened cells, which in the very young Starfish are
prolonged into muscular tails; in the older animal these tails are
separated off as a distinct muscular layer lying between the
ectoderm and the cells lining the cavity of the tube. The tube-foot is
prolonged inwards into a bulb termed the "ampulla," which projects
into the coelom of the arm and in consequence is covered outside by
somatic peritoneum. Just where the ampulla passes into the tube-
foot proper the organ passes downwards between two of the
powerful ambulacral ossicles which support the ambulacral groove,
and a little below this spot a short transverse canal connects the
tube-foot with the radial canal which lies beneath these ossicles (Fig.
191).
Fig. 190.—A, view of the under half of a specimen of Asterias rubens, which has
been horizontally divided into two halves. B, enlarged view of the axial
sinus, stone-canal and genital stolon cut across. amb.oss, Ambulacral
ossicle; amp. ampullae of the tube-feet; ax.s, axial sinus; gon, gonad; g.stol,
genital stolon; marg, marginal ossicle; nerv.circ, nerve ring; oe, cut end of
oesophagus; pst, peristome; ret, retractor muscle of the stomach; sept,
interradial septum; stone c, stone-canal; T, Tiedemann's body; w.v.r, water-
vascular ring-canal.
The tube-feet are, therefore, really a double row of lateral branches

of the radial canal. The appearance of being arranged in four rows is
due to the fact that the transverse canals connecting them with the
radial canal are alternately longer and shorter so as to give room for
more tube-feet in a given length of the arm. Each tube-foot ends in a
round disc with a slightly thickened edge. The radial canal terminates
in a finger-shaped appendage, called the median tentacle, at the
base of which is the eye.
The manner in which this complicated system acts is as follows:—
When the tube-foot is to be stretched out the ampulla contracts and
drives the fluid downwards. The contraction of the ampulla is brought
about by muscles running circularly around it. The tube-foot is thus
distended and its broad flattened end is brought in contact with the
surface of the stone over which it is moving and is pressed close
against it. The muscles of the tube-foot itself, which are arranged
longitudinally, now commence to act, and the pressure of the water
preventing the tearing away of the sucker from the object to which it
adheres, the Starfish is slowly drawn forward, whilst the fluid in the
tube-foot flows back into the ampulla.
Fig. 191.—Diagrammatic cross-section of the arm of a Starfish. adamb,

Adambulacral ossicle; amb, ambulacral ossicle; amp, ampulla of tube-foot;
branch, papula; car, carinal plate; d.lat, dorso-lateral plate; inf.marg, infero-
marginal plate; p.br, peribranchial space; ped, pedicellaria; s.marg, supero-
marginal plate. The nervous ridge between the bases of the tube-feet and
the two perihaemal canals above this ridge are shown in the figure but not
lettered.
If each tube-foot were practically water-tight, then each would be

entirely independent of all the rest, and it would not be easy to
suggest a reason for the presence of the complicated system of
radial canals and stone-canal. Just at the spot, however, where the
transverse canal leading from the radial canal enters the tube-foot
there is a pair of valves which open inwards and allow fluid to pass
from the radial canal into the tube-foot but prevent any passing
outwards in the reverse direction. The presence of these valves
renders it probable that the tube-foot is not quite water-tight; that
when it is distended under the pressure produced by the contraction
of the muscles of the ampulla, some fluid escapes through the
permeable walls; and that the loss thus suffered is made up by the
entry of fresh fluid from the radial canal. The radial canal in turn
draws from the ring-canal, and this last is supplied by the stone-
canal, the cilia of which keep up a constant inward current.
In the fluid contained in the water-vascular system, as in the

coelomic fluid, there are amoebocytes floating about. These are
produced in short pouches of the ring-canal, nine in number, which
are called after their discoverer "Tiedemann's bodies" (Fig. 190, T).
From the cells lining these the amoebocytes are budded off.
The nervous system of the Starfish is in a very interesting

condition. The essential characteristic of all nervous systems is the
presence of the "neuron," a cell primitively belonging to an
epithelium but which generally has sunk below the level of the others
and lies amongst their bases. This type of cell possesses a round
body produced in one direction into a long straight process, the
"axon," whilst in the other it may have several root-like processes, or
"dendrites," which may spring from a common stem, in which case
the neuron is said to be "bipolar." The axon is often distinguished as
a "nerve-fibre" from the round body which is termed the "nerve-cell."
This is due to the fact that for a long time it was not recognised that
these two structures are parts of a whole.
Now at the base of the ectoderm all over the body of the Starfish
there is to be found a very fine tangle of fibrils; these are to be found
partly in connexion with small bipolar neurons lying amongst them
and partly with isolated sense-cells scattered amongst the ordinary
ectoderm cells. This nervous layer is, however, very much thickened
in certain places, so as to cause the ectoderm to project as a ridge.
One such ridge is found at the summit of each ambulacral groove
running along the whole under surface of the arm and terminating in
a cushion at the base of the median tentacle of the water-vascular
system. This ridge is called the radial nerve-cord. The five radial
nerve-cords are united by a circular cord, the nerve-ring, which
appears as a thickening on the peristome surrounding the mouth.
The sense-organs of the Starfish are chiefly the discs of the tube-
feet. Round the edges of these there is a special aggregation of
sense-cells; elsewhere, as in the skin of the back, only isolated
sense-cells are found, and it becomes impossible to speak of a
sense-organ.
A prolongation of the radial nerve-cord extends outwards along one

side of each tube-foot. This is often spoken of as the "pedal nerve,"
but the term nerve is properly retained for a mere bundle of axons
such as we find in the higher animals, whereas the structure referred
to contains the bodies of nerve-cells as well as their outgrowths or
cell-fibres and is therefore a prolongation of the nerve-cord.
Fig. 192.—Diagrammatic longitudinal section through a young Asteroid passing

through the tip of one arm and the middle of the opposite interradius. This
diagram is generalised from a section of Asterina gibbosa. ab, Aboral sinus;
ax, axial sinus; ax1, basal extension of axial sinus forming the inner
perihaemal ring-canal; br, branchia = gill = papula; g.r, genital rachis; mp,
madreporite; musc.tr, muscle uniting a pair of ambulacral ossicles; nerv.circ,
nerve-ring; n.r, radial nerve-cord; oc, eye-pit; oss, ossicles in skin; p.br,
peribranchial sinus; p.c, pore canal; perih (on the right), perihaemal radial
canal, (on the left), outer perihaemal ring-canal; py, pyloric caecum; rect,
rectum; rect.caec, rectal caeca; sp, spines; st.c, stone-canal; t, median
tentacle terminating radial canal; w.v.r, water-vascular radial canal. The
genital stolon (not marked by a reference line) is seen as an irregular band
accompanying the stone-canal, its upper end projects into a small closed
sac, also unmarked, which is the right hydrocoele or madreporic vesicle.
At the base of the terminal tentacle the radial nerve-cord ends in a

cushion. This cushion is called the "eye," for it is beset with a large
number of cup-shaped pockets of the ectoderm. Each pocket is lined
partly by cells containing a bright orange pigment and partly by
visual cells each of which ends in a small clear rod projecting into the
cavity of the pit (Fig. 193, A, vis.r). The pit is apparently closed by a
thin sheet of cuticle secreted by the most superficial cells.
An exposed nervous system and simple sense-organs such as the

Starfish possesses lend themselves admirably to the purposes of
physiological experiment, and so Starfish have been favourite
"corpora vilia" with many physiologists.
Fig. 193.—A, longitudinal section of a single eye-pit of Asterias. s.n, Nucleus of

supporting cell; vis.n, nucleus of visual cell; vis.r, visual rod. B, view of the
terminal tentacle showing the eye-pits scattered over it. (After Pfeffer.)
The light-perceiving function of the eye is easily demonstrated. If a

number of Starfish be put into a dark tank which is illuminated only
by a narrow beam of light they will be found after an interval to have
collected in the space reached by the beam of light.[445] If all the
median tentacles but one be removed this will still be the case; if,
however, they are all removed the Starfish will exhibit indifference to
the light.
If the under surface of a Starfish be irritated by an electric shock or a

hot needle, or a drop of acid, the tube-feet of the affected area will
be strongly retracted, and this irritation will be carried by the pedal
nerves to the radial nerve-cord, with the result that finally all the
tube-feet in the groove will be retracted and the groove closed by the
action of the transverse muscle connecting each ambulacral ossicle
with its fellow. If, on the other hand, the back of a Starfish be irritated
this may produce a contraction of the tube-feet if the irritation be
strong, but this will be followed by active alternate expansions and
contractions, in a word, by endeavours to move. Preyer[446] by
suspending a Starfish ventral surface upward, by means of a small
zinc plate to which a string was attached which passed through a
hole bored in the back and through the mouth, caused movements of
this description which lasted for hours. Irritation of the back causes
also activity of the local pedicellariae, which open their valves widely
and then close them with a snap in the endeavour to seize the
aggressor.
The uninjured Starfish in moving pursues a definite direction, one

arm being generally directed forwards, but this may be any one of
the five. The tube-feet of this arm are directed forwards when they
are stretched out, by the slightly unequal contraction of the
longitudinal muscles of opposite sides of the foot, which persists
even when the circular muscles of the ampulla are contracting. They
thus may be said to swing parallel to the long axis of the arm. The
tube-feet of the other arms assist in the movement, and hence swing
obliquely with reference to the long axis of the arm to which they
belong, although they move parallel to the general direction in which
the Starfish is moving. A change in the direction of the swing of the
tube-feet will bring about a change in the direction of the movement
of the animal as a whole. If now the connexion of each radial nerve-
cord with the nerve-ring be cut through, each arm will act as a
separate Starfish and will move its tube-feet without reference to the
movement of those in the other arms, so that the animal is pulled
first one way and then another according as the influence first of one
arm and then of another predominates. Similarly, when a Starfish is
placed on its back, it rights itself by the combined action of the tube-
feet of all the arms, extending them all as widely as possible, those
which first catch hold being used as the pivot for the turning
movement. If, however, the radial nerve-cords are cut through, each
arm tries to right itself and it is only by chance that the efforts of one
so predominate as to turn the whole animal over. From these
experiments it is clear that the nerve-ring acts as co-ordinator of the
movements of the Starfish, that is to say as its brain.
If a section be taken across the arm of a Starfish (Fig. 191), it will be

seen that between the V-shaped ridge constituting the radial nerve-
cord and the radial water-vascular canal there are two canals lying
side by side and separated from one another by a vertical septum.
These canals are not mere splits in the substance of the body-wall,
but have a well-defined wall of flattened cells. They are termed, for
reasons which will be explained subsequently, perihaemal canals,
and they open into a circular canal called the "outer perihaemal ring,"
situated just beneath the water-vascular ring-canal (Fig. 192, perih).
These canals originate as outgrowths from the coelom. From their
upper walls are developed the muscles which connect the pairs of
ambulacral ossicles and close the groove, and also those which
connect each ossicle with its successor and predecessor and help to
elevate or depress the tip of the arm.
In most of the higher animals the processes of many of the ganglion-

cells are connected together in bundles called "motor nerves," which
can be traced into contact with the muscles, and thus the path along
which the stimulus travels in order to evoke movement can clearly be
seen. No such well-defined nerves can be made out in the case of
the Starfish, and it is therefore interesting when exceptionally the
paths along which stimuli travel to the muscles can be traced. This
can be done in the case of the muscles mentioned above. Whereas
they originate from the dorsal walls of the perihaemal canals,
ganglion-cells develop from the ventral walls of these canals, which
are in close contact with the nerve-cord, so that the nervous system
of the Starfish is partly ectodermic and partly coelomic in origin.
Stimuli reaching the ectodermic ganglion-cells are transmitted by
them to the nervous part of the wall of the perihaemal canal and from
that to the muscular portion of the same layer of cells.
Besides the radial perihaemal canals and their connecting outer

perihaemal ring there are several other tubular extensions of the
coelom found in the body-wall. These are:—
(1) The "inner perihaemal canal," a circular canal in close contact

with the inner side of the outer perihaemal canal (Fig. 192, ax1).
(2) The "axial sinus" (ax) a wide vertical canal embedded in the
body-wall outside the stone-canal. This canal opens into the inner
perihaemal canal below; above it opens into several of the pore-
canals and into the stone-canal. The separation of the axial sinus
from the rest of the coelom is the remains of a feebly marked
metamerism in the larva.
(3) The "madreporic vesicle," a closed sac embedded in the dorsal

body-wall just under the madreporite. This sac by its history in the
larva appears to be a rudimentary counterpart of the water-vascular
system, since this organ in correspondence with the general bilateral
symmetry of the larva is at first paired. Into this a special process of
the genital stolon projects.
(4) The "aboral sinus" (Fig. 192, ab), a tube embedded in the dorsal
body-wall running horizontally round the disc. The aboral sinus
surrounds the genital rachis (see p. 452) and gives off into each arm
two branches, the ends of which swell so as to surround the genital
organs. It has no connexion with the axial sinus though the contrary
has often been stated by Ludwig.[447]
(5) The "peribranchial spaces," circular spaces which surround the

basal parts of the papulae (Fig. 192, p.br).
Besides these, large irregular spaces have been described as

existing in the body-wall by Hamann[448] and other authors, but for
various reasons and especially because they possess no definite
wall they appear to be nothing more than rents caused by the
escape of CO2 gas during the process of decalcifying, to which the
tissues of the Starfish must be subjected before it is easy to cut
sections of them.
The question as to whether or not there is a blood system in the

Starfish has an interesting history. It must be remembered that the
examination of the structure of Echinodermata was first undertaken
by human anatomists, who approached the subject imbued with the
idea that representatives of all the systems of organs found in the
human subject would be found in the lower animals also. So the
perihaemal canals were originally described as blood-vessels. Later,
Ludwig[449] discovered a strand of strongly staining material running
in each septum which separates the two perihaemal canals of the
arm. Each of these radial strands could be traced into connexion
with a circular strand interposed between the outer and the inner
perihaemal ring-canals. This circular strand again came into
connexion with a brown, lobed organ, lying in the wall of the axial
sinus, and this in turn joined at its upper end a circular cord of
pigmented material adhering to the dorsal wall of the coelom (lying in
fact within the aboral sinus), from which branches could be traced to
the generative organs. Ludwig concluded that he had at last
discovered the true blood-vessels, though the facts that the radial
strands and the oral circular strand absorbed neutral carmine
strongly and that the vertical and aboral strands were pigmented,
constituted a very slender basis on which to found such a
conclusion. The colour apparently appealed to the imagination, and it
is undoubtedly true that the "plasma" or blood-fluid of other animals
often absorbs stain strongly.
The strands were accordingly named "radial blood-vessels," "oral

blood-ring," "aboral blood-ring"; and the brown vertical strand was
called the "heart," although no circulation or pulsations had ever
been observed. When later investigations revealed the fact that the
so-called heart was practically solid, the term "central blood-plexus"
was substituted for heart, although it was still regarded as the central
organ of the system. The name "perihaemal" was given to the
spaces so called because they surrounded the supposed blood-
vessels.
In order to come to a satisfactory conclusion on the matter some

general idea as to the fundamental nature and function of the blood-
vessels in general must be arrived at. Investigations made on
various groups of animals, such as Annelida, Mollusca, Crustacea,
Vertebrata, show that at an early period of development a
considerable space intervenes between the alimentary canal and the
ectoderm, which is filled with a more or less fluid jelly. Into this cavity,
the so-called "primary body-cavity" or "archicoel," amoebocytes,
budded from the ectoderm or endoderm or both, penetrate. In this
jelly with its contained amoebocytes is to be found the common
rudiment both of the connective tissue and of the blood system. The
resemblance of the archicoele and its contents to the jelly of a
Medusa is too obvious to require special insistence on, and therefore
in the Coelenterata it may be stated that there is to be found a tissue
which is neither blood system nor connective tissue but is the
forerunner of both.
In the higher animals as development proceeds the jelly undergoes

differentiation, for some of the amoebocytes become stationary and
connected with their pseudopodia so as to form a protoplasmic
network. A portion of this network becomes altered into tough fibres,
but a portion of each strand remains living, and in this way the
connective tissue is formed. In the interstices of the network of fibres
a semi-fluid substance (the unaltered jelly) is found, and this is
traversed by free, wandering amoebocytes. In other places the jelly
becomes more fluid and forms the plasma, or liquid of the blood,
whilst the amoebocytes form the blood corpuscles. The blood
system thus arises from regions of the archicoel where fibres are not
precipitated.
Now in the Starfish the whole substance of the body-wall intervening

between the ectoderm and the coelomic epithelium really represents
the archicoel. The formation of fibres has, it is true, proceeded to a
certain extent, since there are interlacing bundles of these, but there
are left wide meshes in which amoebocytes can still move freely.
Apart from the skeleton, therefore, the tissues of the body-wall of the
Starfish do not exhibit much advance on those of a Jellyfish. If
anything is to be compared to the blood system of the higher animals
it must be these meshes in the connective tissue. From observations
made on other Echinoderms it appears probable that the colour of
the skin is due to amoebocytes loaded with pigment wandering
outwards through the jelly of the body-wall and disintegrating there.
The strands regarded as blood-vessels by Ludwig are specially
modified tracts of connective tissue in which fibres are sparse, and in
which there are large quantities of amoebocytes and in which the
"jelly" stains easily. Cuénot[450] suggests that they are placed where
new amoebocytes are formed; this is quite possible, and in this case
they ought to be compared to the spleen and other lymphatic organs
of Vertebrates, and not to the blood-vessels.[451]
The organ regarded as the heart, however, belongs to a different

category: it is really the original seat of the genital cells and should
be termed the "genital stolon." Careful sections show that at its
upper end it is continuous with a strand of primitive germ-cells which
lies inside the so-called aboral blood-vessel, and is termed the
"genital rachis" (Fig. 192, g.r). The germ-cells are distinguished by
their large nuclei and their granular protoplasm. The genital organs
are only local swellings of the genital rachis, and from the shape of
some of the germ-cells it is regarded as highly probable that the
primitive germ-cells wander along the rachis and accumulate in the
genital organs. The genital rachis itself is an outgrowth from the
genital stolon, and this latter originates as a pocket-like ingrowth of
the coelom into the wall separating it from the axial sinus; when fully
formed it projects into and is apparently contained in this latter
space.
Not all the cells forming the genital stolon become sexual cells. Many
degenerate and become pigment-cells, a circumstance to which the
organ owes its brown colour. In very many species of Starfish many
of the cells of the genital rachis undergo a similar degeneration, and
hence is produced the apparent aboral blood-vessel. Further, the
rachis is embedded in connective tissue which has undergone what
we may call the "lymphatic" modification, and this for want of a better
name we call the "aboral" blood-ring.
The size of the genital organs varies with the season of the year;
they are feather-shaped, and attached to the genital rachis by their
bases, but project freely into the coelom of the arm. From their great
variation in size and also from the shape of some of the cells in the
genital rachis, Hamann concludes that as each period of maturity
approaches fresh germ-cells are formed in the rachis and wander
into the genital organ and grow there in size. It is probable that the
aboral end of the genital stolon is the seat of the formation of new
germ-cells.
In the Starfish, therefore, as in other animals with a well-defined

coelom, the genital cells ultimately originate from the coelomic wall.
The genital ducts are formed by the burrowing outwards of the germ-
cells. When it is remembered that the fundamental substance of the
body-wall is semi-fluid jelly, this process will be better understood.
When the ova and spermatozoa are ripe, they are simply shed out
into the sea and fertilisation occurs there. The development is
described in Chapter XXI. The free-swimming larval period lasts
about six weeks.
Having described a single species with some degree of fulness, we

must now give some account of the range of variation of structure
met with in the group.
Number of Arms.—In the overwhelming majority of Starfish the

number of arms is 5, but deviations from this rule are met with not
only as individual variations, but as the characteristics of species,
genera, and even families.
The number 5 is rarely diminished, but amongst a large collection of

specimens of Asterina gibbosa, belonging to the author, some 4-
rayed individuals are met with. One species of Culcita, C. tetragona,
is normally 4-rayed.
On the other hand the number 5 is often exceeded. The families

Heliasteridae and Brisingidae are characterised by possessing
numerous (19-25) arms. In the normally 5-rayed family Asteriidae
Pycnopodia has 22 arms; and in the Solasteridae the genera
Rhipidaster and Solaster are characterised by possessing 8 and 11-
15 arms respectively; whilst Korethraster and Peribolaster have only
5. The common Starfish of the Gulf of St. Lawrence, Asterias polaris,
is 6-rayed, whilst most of the other species of the same genus are 5-
rayed, though 6 rays are often met with as a variation.
In some species the fact that the number of arms exceeds 5 seems
to be connected with the power of multiplication by transverse
fission. Thus Ludwig[452] has shown that in Asterias tenuispina the
number of arms is usually 7, but sometimes 5, 6, or 8, and that in
most cases the arms are arranged in two groups—one consisting of
small arms, the other of large.
Shape.—Apart from the varying number of arms, differences in the

shape of the Starfish are due to two circumstances:—
(1) The proportion of breadth to length of arm; and
(2) The amount of adhesion between adjacent arms.
The adhesion can go so far that the animal acquires the shape of a
pentagonal disc. This is the case for instance in Culcita. The fact that
the body of this animal is really composed of adherent arms is at
once made clear when the coelom is opened. This space is found to
be divided up by inwardly projecting folds called interradial septa,
which are stiffened by calcareous deposits and represent the
conjoined adjacent walls of two arms.
In the family Heliasteridae the mutual adhesion between the arms

has gone on merely to a slight extent, for the interradial septa are
still double.
Skeleton.—Most of the schemes of classification have been

founded on the skeleton, largely because the greater number of
species have only been examined in the dried condition, and little is
known of their internal anatomy or habits. There is, however, this
justification for this procedure, that the habits and food of the species
(with the exception of the Paxillosa) which have been observed in
the living condition appear to be very uniform, and that it is with
regard to the skeleton that Asteroidea seem to have split into
divergent groups through adopting different means of protecting
themselves from their foes.
The description of the various elements of the skeleton will be

arranged under the following heads:—(a) Main framework; (b)
Spines; (c) Pedicellariae; (d) Ambulacral skeleton.
(a) Main Framework.—The type of skeleton which supports the

body-wall of Asterias is called reticulate. As already indicated it
consists of a series of rods bound together by bundles of connective-
tissue fibres so as to form a mesh-work. This is a very common type
of aboral skeleton, but in a large number of Starfish a different type
occurs, consisting of a series of plates which may fit edge to edge,
leaving between them only narrow interstices, as in the
Zoroasteridae, or which may be placed obliquely (as in Asterina) so
that they imbricate or overlap one another. In a very large number of
Asteroidea the supero- and infero-marginal ossicles are represented
by squarish plates even when the rest of the skeleton is reticulate;

Computer Vision ECCV 2018 15th European Conference Munich Germany September 8 14 2018 Proceedings Part I Vittorio Ferrari

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computer Vision ECCV 2018 15th European Conference Munich Germany September 8 14 2018 Proceedings Part I Vittorio Ferrari

Uploaded by

Copyright:

Available Formats

Computer Vision ECCV 2018 15th

European Conference Munich Germany

Computer Vision ECCV 2018 15th European Conference

Computer Vision ECCV 2018 15th European Conference

Computer Vision ECCV 2018 15th European Conference

Computer Vision ECCV 2018 15th European Conference

Computer Vision ECCV 2018 15th European Conference

Computer Vision ECCV 2018 Workshops Munich Germany

Computer Vision ECCV 2018 Workshops Munich Germany

Computer Vision ECCV 2016 14th European Conference

Cristian Sminchisescu Yair Weiss (Eds.)

ISSN 0302-9743 ISSN 1611-3349 (electronic)

Library of Congress Control Number: 2018955489

© Springer Nature Switzerland AG 2018

September 2018 Horst Bischof

Welcome to the proceedings of the 2018 European Conference on Computer Vision

existing AC assignment, as well as minimizing the fragmentation across ACs, so that

September 2018 Vittorio Ferrari

Program Committee Co-chairs

Local Arrangements Chairs

Industrial Liaison Chairs

Sanja Fidler University of Toronto, Canada

Yasuyuki Matsushita Osaka University, Japan

Tinne Tuytelaars KU Leuven, Belgium

Technical Program Committee

Hassan Abu Alhaija Peter Anderson Arunava Banerjee

Jesus Bermudez-Cameo Xun Cao Erkang Cheng

Achal Dave Aykut Erdem Ryo Furukawa

Venu Govindu Wolfgang Heidrich Evren Imre

Martin Kampel Dimitrios Kosmopoulos Victor Lempitsky

Shih-Yao Lin Simon Lucey Christopher Mei

P. J. Narayanan Hyun Soo Park Victor Prisacariu

Gernot Riegler Torsten Sattler Tianmin Shu

Qing Sun Chetan Tonde Matthias Vestner

Zhaowen Wang Jin Xie Michael Ying Yang

Yang Yu Quanshi Zhang Guang-Tong Zhou

Learning for Vision

Convolutional Networks with Adaptive Inference Graphs . . . . . . . . . . . . . . . 3

Progressive Neural Architecture Search . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Diverse Image-to-Image Translation via Disentangled Representations . . . . . . 36

Lifting Layers: Analysis and Applications . . . . . . . . . . . . . . . . . . . . . . . . . 53

Learning with Biased Complementary Labels . . . . . . . . . . . . . . . . . . . . . . . 69

Semi-convolutional Operators for Instance Segmentation . . . . . . . . . . . . . . . 89

Skeleton-Based Action Recognition with Spatial Reasoning and Temporal

Fictitious GAN: Training GANs with Historical Models . . . . . . . . . . . . . . . 122

Bi-box Regression for Pedestrian Detection and Occlusion Estimation . . . . . . 138

C-WSL: Count-Guided Weakly Supervised Localization . . . . . . . . . . . . . . . 155

Attributes as Operators: Factorizing Unseen

Product Quantization Network for Fast Image Retrieval . . . . . . . . . . . . . . . . 191

Cross-Modal Hamming Hashing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

Deep Video Quality Assessor: From Spatio-Temporal Visual Sensitivity

Semi-dense 3D Reconstruction with a Stereo Event Camera . . . . . . . . . . . . . 242

Self-Calibrating Isometric Non-Rigid Structure-from-Motion. . . . . . . . . . . . . 259

Semi-supervised Deep Learning with Memory . . . . . . . . . . . . . . . . . . . . . . 275

Deep Fundamental Matrix Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

TrackingNet: A Large-Scale Dataset and Benchmark for Object Tracking

StarMap for Category-Agnostic Keypoint and Viewpoint Estimation . . . . . . . 328

Factorizable Net: An Efficient Subgraph-Based Framework for Scene

Multi-fiber Networks for Video Recognition. . . . . . . . . . . . . . . . . . . . . . . . 364

Zero-Shot Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397

A Modulation Module for Multi-task Learning with Applications

Fast and Accurate Intrinsic Symmetry Detection . . . . . . . . . . . . . . . . . . . . . 433

Objects that Sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451