0%
前言
学过 C/C++, 所以最基础部分就跳过啦~
Title : Defending Against Universal Adversarial
Patches by Clipping Feature Norms
Author : Cheng Yu, Jiansheng Chen, Youze Xue, Yuyang
Liu, Weitao Wan, Jiayu Bao, Huimin Ma
Institution :
Department of Electronic Engineering, Tsinghua University,
China
Beijing National Research Center for Information Science and
Technology, China
University of Science and Technology Beijing, China
Year : 2021
Journal : ICCV
Source : Open
access ,Supplemental (注:开源代码在补充材料中
(゜▽゜*) ), IEEE
Cite : Cheng Yu, Jiansheng Chen, Youze Xue, Yuyang
Liu, Weitao Wan, Jiayu Bao, Huimin Ma; Proceedings of the IEEE/CVF
International Conference on Computer Vision (ICCV), 2021, pp.
16434-16442
Idea :
从实验和数学推导证明了对抗补丁会导致特征范数变大并提出了特征范数裁剪的方法来防御对抗补丁攻击
Title : On Feature Normalization and Data
Augmentation
Author : Boyi Li, Felix Wu, Ser-Nam Lim, Serge
Belongie, Kilian Q. Weinberger
Institution : Cornell University, Cornell Tech,
ASAPP, Facebook AI
Year : 2021
Journal : CVPR
Source : Open
access , Arxiv , official PyTorch
implementation
Cite : Boyi Li, Felix Wu, Ser-Nam Lim, Serge
Belongie, Kilian Q. Weinberger; Proceedings of the IEEE/CVF Conference
on Computer Vision and Pattern Recognition (CVPR), 2021, pp.
12383-12392
Idea : 交换 feature map 的矩信息做数据增强
Title : CrossNorm and SelfNorm for Generalization
under Distribution Shifts
Author : Zhiqiang Tang
Institution : Amazon Web Services
Year : 2021
Journal : ICCV
Source : ICCV
2021 open access , arxiv
Code : Official
implementation
Idea : 对 feature map
通道的均值和方差做交换以提高泛化性鲁棒性
Title : When Facial Expression Recognition Meets
Few-Shot Learning: A Joint and Alternate Learning Framework
Author : Xinyi Zou, Yan Yan, Jing-Hao Xue, Si Chen,
Hanzi Wang
Institution : 厦门大学
Year : 2022
Journal : AAAI
Source : Arxiv , PDF
Idea :
利用少样本学习中的孪生网络来解决多源域单表情到多表情的跨域任务,并提出两阶段训练框架提高其泛化性能
Title : Dynamic Convolution: Attention over
Convolution Kernels
Author : Yinpeng Chen, Xiyang Dai, Mengchen Liu,
Dongdong Chen, Lu Yuan, Zicheng Liu
Institution : 微软亚研院
Year : 2022
Journal : CVPR
Source : Arxiv , open
access
Idea :
将多个卷积核通过注意力机制合成一个动态卷积核提高模型的表达能力