University of Science and Technology of China (USTC)
Dong Liu, Associate Professor, University of Science and Technology of China (USTC) His research interests include image and video coding and multimedia data analytics. He received the Ph.D. degree in 2009 from USTC. He was a Member of Research Staff at Nokia Research Center, Beijing, from 2009 to 2012. He joined the School of Information Science and Technology of USTC in 2012. He has co-authored more than 100 papers in journals and international conferences, which were cited more than 900 times. He has 15 granted patents. He received the 2009 IEEE TCSVT Best Paper Award. He and his team won several challenges that were in conjunction with CVPR, ECCV, ACM MM, etc.
Track: Video Codecs
2019/08/24 09:30 - 18:10
Topics in the track
Deep learning has achieved great success in the computer vision and image processing fields, which inspires us to try to apply deep learning into image/video compression in the hope of increasing compression efficiency. In this talk, I shall share with audience some exploration of our research team as well as our perspective at this aspect. My talk will introduce some new techniques in detail, including CNN-based interpolation, compact- and super-resolution-based image/video compression, trained wavelet transform, integer-arithmetic-only deep networks.
With the rapid development of picture services for a large amount of WeChat users, eg. Moments picture, Message picture, picture in WeChat Official Accounts Platform, Sticker, Avatar, etc., WeChat Multimedia Center keeps exploring cutting-edge technologies in video coding and image processing, which has been applied in all kinds of WeChat multimedia services and constantly improved, in order to achieve better image quality at a reasonable operation cost. This presentation will introduce the framework of WeChat self-developed image engine and some key technologies, share the experience in how to develop an efficient image codec for internet image applications, and how to make an image compression strategy to strike a better balance between image quality and bandwidth. Meanwhile, some obscure problems in the practice of WeChat animated and still picture compression as well as the solutions will be discussed.
H.265 software encoder developed by Bilibili and some optimization for offline and online video transcoder
A H.265 software encoder, named as Yhevc, is developed by Bilibili and is used in offline and online video transcoder. Several algorithms are developed to optimize existing pipeline and improve video experience.
Firstly, the characteristics and advantages of AVS2 video coding standard will be introduced. Then, we will share the efficient video encoding algorithms explored by ArcVideo in the process of developing 4K UHD AVS2 Encoder. These algorithms include various fast mode decision strategies. And at last, some experence using GPU to speed up avs2 encoding will be introduced.
Like a fine wine, video codecs get better with age as their reference and commercial encoding performance improves while device support widens. It has only been a few years since AV1’s creation and VVC is even younger, so these codecs are still hardening themselves in the market.This talk will take a look at their early progress from both a technical, adoption and hardware implementation perspective.The preliminary version of VVC is already outperforming HEVC significantly and the goal for the final version is set to reduce the bitrate compared to HEVC by another 40-50% at the same visual quality. Similarly, AV1 is also designed to compete with H.265/HEVC and is targeting an expected improvement of about 30% over VP9/HEVC with only reasonable increases in encoding and playback complexity.So how close are they to actually achieving these goals in the real world? How much are these codecs actually useable in the real world for normal customers?