报告题目:A Novel CNN-Based Zero-Shot Single Image Super-Resolution Network
报告时间:2019年12月29日(周日)下午14:00-15:00
报告地点:北辰威尼斯欢乐娱人v3676楼(西教一)102报告厅
报 告 人:李凌
报告简介:
There have been an increasing number of image super-resolution methods based on deep learning techniques in the recent years, but the majority of them fall into the supervised category, i.e. they need to be trained with substantial amounts of training data and in a lengthy period, which may not always be feasible in practice. Zero-shot unsupervised approach can greatly alleviate this problem. However, very little work in this category exists in the literature. Recently, a Zero-Shot Super-Resolution (ZSSR) method is proposed to generate high-resolution (HR) images from their low-resolution (LR) counterparts. ZSSR employs a convolutional neural network (CNN) to represent transformations from LR images to HR images and is trained only based on a single image. ZSSR achieves the state-of-the-art performance on both real-life LR images and several benchmark datasets. However, the training of the CNN network for ZSSR is not stable since the rectifier is used as the activation function and a custom learning rate adjustment policy is used in conjunction with the Adam optimiser. This means the training for ZSSR could be time-consuming. We propose to use parametric rectifier as the activation function to accelerate the convergence of the network and present an improved algorithm for the training to shorten the training time. The network architecture and the optimisation algorithm used in training are also altered to boost the performance. Experimental results demonstrate that our proposed method outperforms the ZSSR in terms of both reconstruction accuracy and speed on three benchmark datasets: Set5, Set14, and Urban100, respectively. The proposed method also produces more visually pleasing results on various LR images obtained in real life, e.g. historical images and images taken with a mobile phone.
报告人简介:
Ling Li obtained her Bachelor of Science (Computer Science) from Sichuan University,China, Master of Electrical Engineering from China Academy of Post and Telecommunication, and PhD of Computer Engineering from Nanyang Technological University (NTU), Singapore. She worked as an Assistant Professor and subsequently an Associate Professor in the School of Computer Engineering in NTU, before moving to Curtin University in Perth, Australia. She served as the Head of Department of Computing, and Director of Digital Ecosystem and Business Intelligent Institute. She is now a Professor and the Head of the School of Electrical Engineering, Computing and Mathematical Sciences in Curtin University. Her research interest is mainly in machine learning, computer vision and graphics, and artificial intelligences. She has given a number of keynotes speeches in international conferences and published over 170 referred research papers in international journals and conferences. She has served in many professional bodies such as SIGRRAPH Australia-New Zealand Chapter, WA ICT Skill Leaderships Group, and Standard Australia Committee etc., and sits in the editorial boards of 5 international journals.
热烈欢迎广大教师及研究生参加!