Speaker: Dr. Jun Liu (刘君博士)
Time: 12:30-13:30, 9 November 2021 (Tuesday) (Beijing time)
Venue: C404, Lijiao Building, BNU at Zhuhai
Abstract
The activation functions in the existing network architecture of CNNs are always given and lack capabilities to handle important spatial information in a way that have been done for many well-known traditional variational models. Priors such as nonlocal regularization cannot be well handled by existing CNN architectures. We propose a novel Nonlocal Soft Threshold Dynamics (NLSTD) based activation layers which can easily integrate many priors such as nonlocal and edges information of the classic variational models into the DCNNs for image segmentation. The novelty of our method is to interpret the activation functions (including softmax, sigmoid, ReLU) as primal-dual variational problem, and thus many priors can be imposed in the dual space. By unrolling method, we can build several NLSTD based network architectures which can enable the outputs of CNN to have nonlocal priors. We will give some applications to show the efficiency of our method.
About Dr. Liu
刘君,北京师范大学副教授,博士生导师。2011年博士毕业于北师大。曾受邀访问过美国UCLA、新加坡南洋理工、香港科技大学、香港浸会大学等高校。主要研究方向为变分法及深度学习相关的图像处理算法与应用。一些研究结果发表在图像处理与计算机视觉相关领域国际知名期刊如Int. J. Comput. Vis., IEEE T. Image. Process., IEEE T. Geosci. Remote, Pattern Recogn., SIAM J. Imaging Sci., J. Sci. Comput., J. Math. Imaging Vis. 等。研究成果曾获教育部高等学校优秀科研成果二等奖(团体), 北京市科技进步二等奖(团体)。主持参与多项国家科研项目。
