The result of the CLIC (Challenge of Learned Image Compression) of CVPR (Conference on Computer Vision and Pattern Recognition) 2018 was released. Professor Chen Zhenzhong from School of Remote Sensing and Information Engineering, teaming up with the Silicon Valley Research Center of Tencent Audio and Video Laboratory, won the first prize of the PSNR (Peak Signal to Noise Ratio) in the CLIC.
CVPR is a top-notch conference in the field of computer vision. This year, Google, Twitter, Amazon, and other companies co-initiated the first workshop and challenge of image compression in the CVPR, which is also the first of its kind launched by a conference of computer vision. The newly started challenge is aimed at incorporating neural network, deep learning, and other new technologies into the filed of image compression.
Image compression is of great importance for information transition on the Internet. For example, an uncompressed image of 12 million px will consume a memory of 36MB. And the amount of images transmitted and stored everyday is calculated by trillions. Therefore, a highly efficient algorithm of image compression is necessary for saving broadband and storage resources, and lifting pressures for servers.
According to CVPR, this Challenge evaluated the performance of participants from 2 perspectives, namely PSNR and subjective assessment. PSNR is the error of per-pixel statistics between the before-and-after compressed image. The higher the PSNR is, the smaller the error will be, and the compressed imaged will be more similar to the original one which indicates a smaller loss of image quality. In the process of international standard setting of image and audio, PSNR has been adopted as one of the index.
Finally, iipTiramisu, co-organized by the team of Professor Chen Zhenzhong and Silicon Valley Research Center of Tencent Audio and Video Laboratory, ranked first in PSNR, and also topped the list of other indexes.
Chen Zhenzhong said that with the uncertainty-based resource allocation strategy, iipTiramisu employed coding modules of CNNMC and CNN in-loop filter which are based on traditional hybrid image coder. It helped the team to achieve a 30% higher DS compression quality than BPG, one of the best open-sourcing image compression algorithms.
Chen Zhenzhong have long been committed to computer vision, video and audio processing, man-machine interaction, data mining and other related fields. In recent years, the team of Chen Zhenzhong and Tencent Audio and Video Laboratory have conducted in-depth cooperation in such areas as video and audio processing and AI.
(Rewritten by Li Xin, Edited by Shen Yuxi, Liu Xiaoli)
