4 PoemK

尚未进行身份认证

暂无相关简介

等级
TA的排名 3k+

Layer Normalization

论文: Jimmy Lei Ba, Jamie Ryan Kiros, Geoffrey E. Hinton. Layer Normalization. arXiv:1607.06450Batch normalization对每个channel或者hidden unit,求输入(mini-)batch上所有样本在这个channel上的均值和标准差。之后根据每个channel上的均值和...

2019-09-15 13:44:01

Transferrable Prototypical Networks for Unsupervised Domain Adaptation (CVPR 2019)

Transferrable Prototypical Networks for Unsupervised Domain Adaptation论文: Yingwei Pan, Ting Yao, Yehao Li, Yu Wang, Chong-Wah Ngo, Tao Mei. Transferrable Prototypical Networks for Unsupervised Domain...

2019-09-11 19:58:09

Progressive Feature Alignment for Unsupervised Domain Adaptation-CVPR2019.md

Progressive Feature Alignment for Unsupervised Domain Adaptation (CVPR2019)提出了一个逐步特征对齐网络 Progressive Feature Alignment Network (PFAN)去解决原域有标签、目标域无标签的无监督domain adaptation分类问题:Easy-to-Hard 迁移策略 (EHT...

2019-09-09 16:00:40

Face photo recognition using sketch (人脸画像合成)

paper(原文): Xiaoou Tang, Xiaogang Wang. “Face photo recognition using sketch.” Proceedings. International Conference on Image Processing. Vol. 1. IEEE, 2002.1. eigenface1.1利用协方差矩阵得到eigenvectors(也称为e...

2019-09-03 23:20:23

SSIM(structural similarity index) ---图像质量评价指标之结构相似性

Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. Image Quality Assessment: From Error Visibility to Structural Similarity. IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 13, NO. 4, APRI...

2019-09-01 12:19:11

Universal Style Transfer via Feature Transforms (WCT,风格迁移,NIPS2017)

Li Y, Fang C, Yang J, et al. Universal Style Transfer via Feature Transforms. NIPS 2017风格迁移的关键问题是如何提取有效果的风格特征并且让输入的内容图像去匹配这种风格。前人的工作证明了协方差矩阵和Gram矩阵能较好地反映视觉风格。基于优化的风格迁移方法,可以处理任意风格并且达到满意的效果但是计算代价太大(时间...

2019-08-20 15:55:13

Diversified texture synthesis with feed-forward networks (纹理生成、风格迁移,CVPR2017)

paper: Li, Yijun, et al. “Diversified texture synthesis with feed-forward networks.” IEEE CVPR 2017.目前的判别和生成模型在纹理合成方面有较好的效果。但是,现存的基于前馈神经网络的方法往往牺牲generality(普遍性)来换取效率,这往往会引发以下问题: 1) 训练出的网络缺少generality...

2019-08-17 14:24:44

Adaptive Instance Normalization (AdaIN Normalization) ICCV 2017

paper: Huang, Xun, and Serge Belongie. “Arbitrary style transfer in real-time with adaptive instance normalization.” ICCV 2017.论文首先回顾了Batch Normalization和Instance Normalization1. Batch Normalization...

2019-08-11 21:54:43

A Content Transformation Block for Image Style Transfer (CVPR2019, 风格迁移)

Kotovenko,Dmytro,etal.“AContentTransformationBlockforImageStyleTransfer.”CVPR2019.风格迁移问题:X→YX\rightarrowYX→Y。源域为XXX,目标域为YYY本文中EEE为Encoder,主要负责提取图像内容信息,换句话说也就是去风格化。DDD为Decoder,主要负责将...

2019-08-07 21:51:07

Unsupervised pixel-level domain adaptation with generative adversarial networks (DA+ 图像转换)

**Bousmalis, Konstantinos, et al. “Unsupervised pixel-level domain adaptation with generative adversarial networks.” CVPR2017. **问题背景:将原域的图像转换到目标域。已知源域的图像类别标签,对于目标域的图像标签未知。损失函数1. 域对抗损失:2. 分类损失...

2019-08-05 16:27:14

Discriminative deep metric learning for face verification in the wild 度量学习(CVPR2014)

Hu, Junlin, Jiwen Lu, and Yap-Peng Tan. “Discriminative deep metric learning for face verification in the wild.” CVPR 2014. (度量学习用于Face Verification:)马氏距离:其中MMM是一个d×dd \times dd×d的半正定矩阵,根据Choleskey...

2019-08-02 22:35:54

马氏距离

1. 协方差矩阵是半正定矩阵对于向量xxx,设其均值为uuu。那么协方差矩阵Σ=E[(x−u)(x−u)T]\Sigma=E[(x-u)(x-u)^T]Σ=E[(x−u)(x−u)T]yTΣy=yTE[(x−u)(x−u)T]y=E[yT(x−u)(x−u)Ty]=E[((x−u)Ty)T(x−u)Ty]=E[∣∣(x−y)Ty∣∣2]≥0\begin{aligned}y^T\Sigma ...

2019-08-02 13:48:52

Unsupervised domain adaptation with residual transfer networks(NIPS 2016)

Long, Mingsheng, et al. “Unsupervised domain adaptation with residual transfer networks.” Advances in Neural Information Processing Systems. 2016.问题:domain adaptation用于分类问题。其中source domain具有label,ta...

2019-07-15 11:09:32

Deep coral: Correlation alignment for deep domain adaptation. ECCV 2016. Domain Adaptation

** Sun, Baochen, and Kate Saenko. “Deep coral: Correlation alignment for deep domain adaptation.” ECCV. Springer, Cham, 2016. **结构如图:两个损失函数:其中LCLASS\mathcal{L}_{CLASS}LCLASS​为分类损失,LCORAL\mathcal{...

2019-07-14 19:15:58

Domain Separation Networks (NIPS 2016)

Bousmalis, K., Trigeorgis, G., Silberman, N., Krishnan, D., & Erhan, D. (2016). Domain separation networks. NIPS 2016.网络结构:输入图像为xxx。对于target domain 有两个特征提取网络: Ec(x),Eps(xs)E_c(x),E_p^s(x^s)Ec​...

2019-07-12 22:36:22

Beyond sharing weights for deep domain adaptation (PAMI 2018) ---Transfer Learning

Rozantsev, Artem, Mathieu Salzmann, and Pascal Fua. “Beyond sharing weights for deep domain adaptation.” IEEE transactions on pattern analysis and machine intelligence 41.4 (2018): 801-814. (Domain Ad...

2019-07-09 14:39:54

Deep Domain Confusion: Maximizing for Domain Invariance

Tzeng E , Hoffman J , Zhang N , et al. Deep Domain Confusion: Maximizing for Domain Invariance[J]. Computer Science, 2014.主要使用两个损失函数:1) 对于source domain上(默认有label)数据 (以及target domain上有label的数据)的分类误差进...

2019-07-07 22:08:54

笔记: Gradient Reversal Layer (unsupervised domain adaptation by backpropagation. ICML 2015)

paper:Ganin,Yaroslav,andVictorLempitsky.“Unsuperviseddomainadaptationbybackpropagation.”ICML37.JMLR.org,2015.论文用**domainadaptation**算法解决目标域无标签的分类问题。文章假设sourcedomain有数据xxx,和labelyyy...

2019-07-06 22:57:30

PyTorch 模型参数和optimizer

网络参数(parameters()和named_parameters())model为网络,打印层结构:ret=[*model.modules()]forlayerinret:#print(type(layer))print(layer.__class__)大概会输出这样的一个效果:<class‘main.ResNet’>...

2019-04-11 15:49:08

pandas读取csv并绘制散点图

读取loc='checkpoints/mnist_vae2/mu_and_sigma.csv'importpandasaspdimportnumpyasnpX=pd.DataFrame.from_csv(loc)#X=X.sort_index(by='label')按照label排序X=X.values#转换成熟悉的numpy绘制importmatpl...

2019-03-20 14:37:13

查看更多

勋章 我的勋章
  • 专栏达人
    专栏达人
    授予成功创建个人博客专栏的用户。专栏中添加五篇以上博文即可点亮!撰写博客专栏浓缩技术精华,专栏达人就是你!
  • 持之以恒
    持之以恒
    授予每个自然月内发布4篇或4篇以上原创或翻译IT博文的用户。不积跬步无以至千里,不积小流无以成江海,程序人生的精彩需要坚持不懈地积累!
  • 勤写标兵Lv1
    勤写标兵Lv1
    授予每个自然周发布1篇到3篇原创IT博文的用户。本勋章将于次周周三上午根据用户上周的博文发布情况由系统自动颁发。