Signal Processing - November 2017 - 48

Suha Kwak (skwak@dgist.ac.kr) re--
through selecting those belonging to tarIt will be an interesting
get semantic categories [9], [18], [24], [35].
ceived
the B.S. and Ph.D. degrees from the
approach to weakly superAlso, in [18] and [49], optical flows help
Department of Computer Science and
vised semantic segmentasynthesize more accurate segmentation
Engineering at POSTECH, South Korea, in
tion to newly introduce
annotations from videos by propagating
2007 and 2014, respectively. He is an assisother unsupervised techclass-localization information between
tant professor in the Department of Inforconsecutive frames along with motion. It
mation and Communication Engineering at
niques that can recoup the
will be an interesting approach to weakly
Daegu Gyeongbuk Institute of Science and
gap between the level of
supervised semantic segmentation to
Technology (DGIST), South Korea. Be--
supervision and that
newly introduce other unsupervised techfore joining DGIST, he was a postdoctorof prediction.
niques that can recoup the gap between the
al fellow of the WILLOW team at Inria
level of supervision and that of prediction.
Paris and École Normale Supérieure,
Another direction is transfer learning. Existing benchFrance. His research interests include computer vision and
marks for semantic segmentation [11], [29] provide segmenmachine learning.
tation annotations, and it is obviously desirable to exploit the
Bohyung Han (bhhan@postech.ac.kr) received the B.S.
existing segmentation annotations although they are given for
and M.S. degrees in computer engineering in 1997 and
only a small number of semantic categories that may not be
2000, respectively, from Seoul National University, South
relevant to target categories we would like to segment. TransKorea, and the Ph.D. degree in computer science in 2005
fer learning realizes this motivation by transferring segmenfrom the University of Maryland at College Park. He is curtation knowledge learned for certain categories into that for
rently an associate professor in the Department of Computer
the other categories. Then, from our point of view, the key
Science and Engineering at POSTECH, South Korea. He is
determinants of success in this line of research are network
an associate editor of Computer Vision and Image Underarchitecture and learning strategy that enable DCNNs to
standing and Machine Vision and Applications. He served or
learn segmentation knowledge that can be applied to arbiwill serve as an area chair of the International Conference
trary semantic categories out of the training data set. A pioon Computer Vision (ICCV) 2015/2017, IEEE Conference
neer study has been done by Hong et al. [17], but there is
on Computer Vision and Pattern Recognition 2017, and Annustill much room for improvement in terms of both network
al Conference on Neural Information Processing Systems 2015,
architecture and learning strategy.
and as a tutorial chair in ICCV 2019. His current research interThe aforementioned suggestions are mainly for weakly
ests include computer vision and machine learning with an emsupervised learning of DCNNs for semantic segmentation, but
phasis on deep learning.
we believe that some of the ideas and techniques can be applied
to weakly supervised learning of other complicated visual recReferences
ognition models for which annotated training examples are not
[1] B. Alexe, T. Deselaers, and V. Ferrari, "Measuring the objectness of image
windows," IEEE Trans. Pattern Anal. Mach. Intell., vol. 34, no. 11, pp. 2189-
sufficiently given.
2202 2012.

Acknowledgments
This work was supported in part by the Institute for Information and Communications Technology Promotion grant (20160-00563, Research on Adaptive Machine Learning Technology
Development for Intelligent Autonomous Digital Companion),
National Research Foundation grant (NRF-2011-0031648,
Global Frontier R&D Program on Human-Centered Interaction
for Coexistence), and DGIST R&D Program (17-ST-02), which
are funded by the Korean government. It is also partly supported
by the Ministry of Culture, Sports, and Tourism of Korea, and
Korea Creative Content Agency.

Authors
Seunghoon Hong (maga33@postech.ac.kr) received the B.S.
and Ph.D. degrees from the Department of Computer Science
and Engineering at POSTECH, Pohang, South Korea, in 2011
and 2017, respectively. He is currently a postdoctoral fellow in
the Department of Electrical Engineering and Computer
Science at the University of Michigan. His current research
interests include computer vision and machine learning. He
received the Microsoft Research Asia Fellowship in 2014.
48

[2] P. Arbeláez, J. Pont-Tuset, J. Barron, F. Marques, and J. Malik, "Multiscale
combinatorial grouping," in Proc. IEEE Conf. Vision and Pattern Recognition,
2014, pp. 328-335.
[3] D. Barnes, W. Maddern, and I. Posner, "Find your own way: Weaklysupervised segmentation of path proposals for urban autonomy," in Proc. IEEE
Int. Conf. Robotics and Automation, 2017, pp. 203-210.
[4] A. Bearman, O. Russakovsky, V. Ferrari, and L. Fei-Fei, "What's the point:
Semantic segmentation with point supervision," in Proc. European Conf.
Computer Vision, 2016, pp. 549-565.
[5] L.- C. Chen, G. Papa nd reou, I. Kok k inos, K. Mu r phy, a nd A. L.
Yuille, "Semantic image segmentation with deep convolutional nets and
fully connected CRFs," in Proc. Int. Conf. Learning Representations,
2015.
[6] Y. Chen, L.-C. Yang, J. Wang, W. Xu, and A. L. Yuille, "Attention to scale:
Scale-aware semantic image segmentation," in Proc. IEEE Conf. Computer
Vision and Pattern Recognition, 2016, pp. 3640-3649.
[7] X. Chu, W. Ouyang, H. Li, and X. Wang, "Structured feature learning for pose
estimation," in Proc. IEEE Conf. Computer Vision and Pattern Recognition,
June 2016, pp. 4715-4723.
[8] M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benenson, U.
Franke, S. Roth, and B. Schiele, "The cityscapes dataset for semantic urban scene
understanding," in Proc. IEEE Conf. Computer Vision and Pattern Recognition,
2016, pp. 3213-3223.
[9] J. Dai, K. He, and J. Sun, "BoxSup: Exploiting bounding boxes to supervise
convolutional networks for semantic segmentation," in Proc. IEEE Int. Conf.
Computer Vision, 2015, pp. 1635-1643.
[10] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, "ImageNet: A
large-scale hierarchical image database," in Proc. IEEE Conf. Computer Vision
and Pattern Recognition, 2009, pp. 248-255.

IEEE SIGNAL PROCESSING MAGAZINE

November 2017

Table of Contents for the Digital Edition of Signal Processing - November 2017