A Decomposition Method of Object Transfiguration

Speaker(s): Seung Joon Lee, Sogang University, South Korea
Keon-Woo Kang, Sogang University, South Korea
Siyeong Lee, NAVER LABS, South Korea
Suk-Ju Kang, Sogang University, South Korea

Seung Joon Lee received the B.S degree in electronics engineering from Sogang University, Seoul, South Korea, in 2018, where he is currently pursuing the M.S. degree. His current research interests include computer vision, image processing, and deep learning.

Keon-Woo Kang received the B.S degree in electronics engineering from Sogang University, Seoul, South Korea, in 2018, where he is currently pursuing the M.S. degree. His current research interests include efficient hardware design for deep learning acceleration, computer vision, image processing, and deep learning.

Siyeong Lee received a B.S. degree in Computer Science and Engineering (2017) and an M.S. degree in Electronic Engineering (2019) from Sogang University, Rep. of Korea. He is currently a researcher at NAVER LABS, Rep. of Korea. His current research interests include computer vision, image processing, deep learning and information theory.

Suk-Ju Kang received the B.S.degree in electronics engineering from Sogang University, Seoul, South Korea, in 2006, and the Ph.D.degree in electrical and computer engineering from Pohang University of Science and Technology, Pohang, South Korea, in 2011. From 2011 to 2012, he was a Senior Researcher with LG Display, Seoul, where he was a Project Leader for resolution enhancement and multiview 3D system projects. From 2012 to 2015, he was an Assistant Professor of Electrical Engineering with Sogang University. Research interests include image analysis and enhancement, video processing, multimedia signal processing, and circuit design for LCD, OLED, 3D display systems.

Description: The method that decomposes an object transfiguration task into two subtasks is proposed:object removal and object synthesis. We formulate each task distinguishing a background and an object using instance information.

ChinaStyle: A Mask-Aware Generative Adversarial Network for Chinese Traditional Image Translation

Speaker(s): Yuan Wang, Institute of Information Engineering, Chinese Academy Of Sciences, China
Weibo Zhang, Institute of Information Engineering, Chinese Academy Of Sciences; School of Cyber Security, University of Chinese Academy of Sciences, China
Peng Chen, Institute of Information Engineering, Chinese Academy Of Sciences; School of Cyber Security, University of Chinese Academy of Sciences, China

I am in Institute of Information Engineering，Chinese Academy of Sciences now. I obtained my Master degree from Shandong University in 2017. My research interests are image processing and computer graphics.

I am a third-year doctor of the School of Cyberspace and Security at the University of Chinese Academy of Sciences. I received a master's degree in electronic science and technology from Beijing University of Posts and Telecommunications in 2017 and a bachelor's degree in electronic information science and technology from Lanzhou University in 2011. My research interests include GANs, style transfer, multi-label image classification and so on.

My name is Peng Chen. I am a PhD student of Institute of Infomation Engineering, Chinese Academy of Sciences. I got my undergrad degree in computer sciences from Shandong University in 2016.My research interests is deep generative model，ranging from theory to application. Besides, I also do some works on how to distinguish whether an image or video is generated by deep generation model.

Description: This paper introduced MA-GAN to generate Chinese traditional paintings from portrait photos with our training dataset. The experiment shows it is more effective than existing methods both quantitatively and qualitatively.

Structure-Aware Image Expansion with Global Attention

Speaker(s): Dewen Guo, Peking University, China
Jie Feng, Peking University, China
Bingfeng Zhou, Peking University, China

Dewen Guo is a graduate student of Computer Science in Peking University (PKU). He received his Bachelor degree of Computer Science from East China Normal University (ECNU) in 2018. His research interests include image processing, deep learning and robotics.

Jie Feng received her Bachelor degree from School of Mathematics Science, Peking University, in 2000, and Ph. D. degree of Engineering from Center for Information Science, Peking University, in 2005. Her research interests include image-based rendering, virtual reality, 3D modeling, digital geometry processing, etc.

Bingfeng Zhou received his Ph. D. degree in Peking University. He is now a researcher and doctoral supervisor in Wangxuan Institute of Computer Technology, Peking University. His research interests include virtual reality, color image processing, digital image halftoning, digital geometry processing, image based rendering, none photorealistic rendering, solid modeling, GPU technology and digital video processing.

Description: We propose a learning-based method combining structure-aware and visual attention strategies to for image expansion. We apply our method on a human face dataset, which containing semantic and structural details.

The Power of Box Filters: Real-time Approximation to Large Convolution Kernel by Box-filtered Image Pyramid

Speaker(s): Tianchen Xu, Advanced Micro Devices, Inc. (AMD), China
Xiaohua Ren, Tencent Youtu X-Lab, China
Enhua Wu, State Key Lab of CS, Institute of Software, Chinese Academy of Sciences & Univ. of CAS; University of Macau, China

Tianchen Xu received his BSc degree and MSc degree in Software Engineering from University of Macau in summers of 2011 and 2014 respectively. He was a research assistant in HKUST VisLab in 2015, and then worked for NVIDIA as a developer technology engineer from 2015 to 2016. He is currently a senior software development engineer working for AMD. His research interests include real-time rendering, character animation, and fluid simulation.

Xiaohua Ren is currently a researcher at the Tencent Youtu X-Lab. He obtained his PhD degree from the University of Macau in 2019. His research interests lie in fluid simulation, image/geometry processing, and deep learning.

Enhua Wu received his BSc in 1970 from Tsinghua University and stayed teaching there until 1980. He was awarded PhD degree from University of Manchester in 1984, followed by working at the State Key Lab. of Computer Science, Chinese Academy of Sciences since 1985, and also as a full professor of University of Macau since 1997. He is an Associate Editor-in-Chief of the JCST since 1995, and the editorial board member of CAVW, Visual Informatics. He is a fellow member of the China Computer Federation (CCF). His research interests include realistic image synthesis, physically based simulation, and virtual reality.

Description: This paper presents a novel solution for approximations to some large convolution kernels by leveraging a weighted box-filtered image pyramid set. Our algorithm is kernel-size independent, and fast on GPU.