- 博客(62)
- 资源 (4)
- 收藏
- 关注
原创 Package inputenc Error: Unicode char \uFFFC
Package inputenc Error: Unicode char \uFFFCPackage inputenc Error: Unicode char \uFFFCPackage inputenc Error: Unicode char \uFFFCWhen trying to use \printbibliography in LaTeX document of article wi...
2018-11-08 11:38:12
2416
1
原创 The GAN Landscape: Losses, Architectures, Regularization, and Normalization
The GAN Landscape: Losses, Architectures, Regularization, and NormalizationKarol Kurach, Mario Lucic, Xiaohua Zhai, Marcin Michalski, Sylvain GellyAbstractGAN: successful; notoriously challengin...
2018-07-17 10:50:21
1084
原创 ELEGANT: Exchanging Latent Encodings with GAN for Transferring Multiple Face Attributes
ELEGANT: Exchanging Latent Encodings with GAN for Transferring Multiple Face AttributesTaihong Xiao, Jiapeng Hong, and Jinwen MaAbstracttask: face attribute transfer existing method: image-to-i...
2018-07-13 11:12:32
869
原创 Finding Tiny Faces in the Wild with Generative Adversarial Network
Finding Tiny Faces in the Wild with Generative Adversarial NetworkYancheng Bai, Yongqiang Zhang, Mingli Ding, Bernard GhanemAbstracttask: detecting small faces in unconstrained conditions chall...
2018-07-11 23:38:15
3579
7
原创 A LEARNED REPRESENTATION FOR ARTISTIC STYLE
A LEARNED REPRESENTATION FOR ARTISTIC STYLEVincent Dumoulin, Jonathon Shlens, Manjunath Kudlur ICLR 2017Abstractconstruct a single, scalable deep network that can parsimoniously capture the a...
2018-07-11 15:34:48
812
原创 NICE: NON-LINEAR INDEPENDENT COMPONENTS ESTIMATION
NICE: NON-LINEAR INDEPENDENT COMPONENTS ESTIMATIONLaurent Dinh, David Krueger, Yoshua BengioAbstractmodeling complex high-dimensional densities basic assumption: a good representation is one in...
2018-07-10 22:56:05
1257
3
原创 Glow: Generative Flow with Invertible 1x1 Convolutions
Glow: Generative Flow with Invertible 1××\times1 ConvolutionsDiederik P. Kingma, Prafulla DhariwalAbstractflow-based generative models: tractability of the exact log-likelihood, tractability of ...
2018-07-10 17:08:09
2955
1
原创 UNSUPERVISED CROSS-DOMAIN IMAGE GENERATION
UNSUPERVISED CROSS-DOMAIN IMAGE GENERATIONYaniv Taigman, Adam Polyak & Lior WolfAbstracttransferring a sample in one domain to an analog sample in another domain learn a generative function...
2018-07-10 10:07:16
616
原创 Taskonomy: Disentangling Task Transfer Learning
Taskonomy: Disentangling Task Transfer LearningAmir R. Zamir, Alexander Sax, William Shen, Leonidas Guibas, Jitendra Malik, Silvio Savarese(Computer Vision: from 3D reconstruction to recognition (CS...
2018-06-29 11:10:34
1190
原创 Fully Convolutional Adaptation Networks for Semantic Segmentation
Fully Convolutional Adaptation Networks for Semantic SegmentationYiheng Zhang, Zhaofan Qiu, Ting Yao, Dong Liu, and Tao MeiAbstract问题:语义分割的标注极其困难 一种思路:render synthetic data (e.g., computer game...
2018-06-28 22:57:48
629
原创 YOLOv3: An Incremental Improvement
YOLOv3: An Incremental ImprovementJoseph Redmon Ali Farhadi 语言风趣幽默,一点也不严肃呆板Abstract上一个版本YOLO的“一点”改进Introduction毕竟大牛,几乎玩了一年twitter,剩下的时间改进了YOLOThe DealBounding Box Predictionpred...
2018-06-28 17:33:19
202
原创 Neural scene representation and rendering
Neural scene representation and renderingDeep Mind! S. M. Ali Eslami, Danilo Jimenez Rezende, Frederic Besse, Fabio Viola, Ari S. Morcos, Marta Garnelo, Avraham Ruderman, Andrei A. Rusu, Ivo Danihe...
2018-06-27 08:30:16
1015
原创 Attention Is All You Need
Attention Is All You NeedAbstract任务:机器翻译 传统方法:RNN/CNN,可能加上attention机制 本文的方法:Transformer(变形金刚?变压器?),只用attentionIntroductionRNN’s inherent sequential nature precludes parallelization within tr...
2018-06-25 16:16:22
196
原创 Graphical Generative Adversarial Networks
Graphical Generative Adversarial NetworksChongxuan Li, Max Welling, Jun Zhu, Bo ZhangAbstractGraphical GAN, 建模图像的结构信息(最近的好几篇工作都是围绕这个问题) 两种训练方法:global algorithm treats all variables as a wholel,...
2018-06-25 11:23:21
767
原创 DensePose: Dense Human Pose Estimation In The Wild
DensePose: Dense Human Pose Estimation In The WildRıza Alp G¨uler, Natalia Neverova, Natalia Neverova DensePose-COCO: a large-scale ground-truth dataset with image-to-surface correspondences manu...
2018-06-24 22:53:35
2100
原创 Image-to-Image Translation with Conditional Adversarial Networks
Image-to-Image Translation with Conditional Adversarial NetworksPhillip Isola, Jun-Yan Zhu, Tinghui Zhou, Alexei A. EfrosIntroduction任务:image-to-image translation(translating one possible repr...
2018-06-21 19:39:54
410
原创 Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial NetworksJun-Yan Zhu, Taesung Park, Phillip Isola, Alexei A. EfrosAbstract任务:Image-to-image translation(learn the mapping ...
2018-06-21 16:03:41
817
原创 CartoonGAN: Generative Adversarial Networks for Photo Cartoonization
CartoonGAN: Generative Adversarial Networks for Photo CartoonizationYang Chen, Yu-Kun Lai, Yong-Jin LiuAbstract类似新海诚的《你的名字》,作者做的是transforming photos of real-world scenes into cartoon style image...
2018-06-21 10:43:08
4130
7
原创 Generative Adversarial Network Architectures For Image Synthesis Using Capsule Networks
Generative Adversarial Network Architectures For Image Synthesis Using Capsule Networks类似于SAGAN的出发点,传统CNN仅仅考虑图像的局部信息,没有考虑全局范围内不同区域的联系,而capsule是可以建模图像的空域相关性的,因此用capsule network代替CNN作为判别器,使生成的图像具有结构性。...
2018-06-20 16:25:50
332
原创 Dynamical Isometry and a Mean Field Theory of CNNs
Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla Convolutional Neural NetworksLechao Xiao 1 2 Yasaman Bahri 1 2 Jascha Sohl-Dickstein 1 Samuel S. Schoenholz 1 Je...
2018-06-20 11:09:05
901
原创 Self-Attention Generative Adversarial Networks
Self-Attention Generative Adversarial NetworksHan Zhang(Rutgers University), Ian Goodfellow(Google Brain), Dimitris Metaxas(Rutgers University),Augustus Odena(Google Brain)引言任务:图像生成/图像合成 问题:传统方...
2018-06-10 19:30:15
1665
原创 Arbitrary Style Transfer with Deep Feature Reshuffle
Arbitrary Style Transfer with Deep Feature ReshuffleShuyang Gu, Congliang Chen, Jing Liao, Lu Yuan University of Science and Technology of China, Peking University, Microsoft Research https://arxi...
2018-05-14 16:15:30
1368
原创 模式识别
模式识别主讲老师:汪增福 教材:《模式识别》,汪增福绪论模式:时间和空间中可观测事物的集合,具有可观测性、可区分性和相似性 模式识别:通过对观测样本的分析完成对输入模式的分类进而给出输入模式的描述(本质上就是分类) 模式识别系统: 被测对象∈现实世界−→−−−−模式采集带噪模式∈模式空间−→−−−预处理净化模式∈模式空间−→−−−−−−−−特征提取与表达特征∈特征空间−→−−−...
2018-05-11 22:25:39
1641
原创 计算机视觉
计算机视觉主讲老师:曹洋 课程网站:http://home.ustc.edu.cn/~yzc101/ 教材:Richard Szeliski,Computer Vision: Algorithms and Applications,Springer,2010 考核方式:阅读报告或项目报告绪论研究内容视觉基础⇒底层处理(图像处理、特征提取)⇒中层处理(图像分割、相机...
2018-05-08 14:53:24
461
原创 Lecture 13: Generative Models
CS231nLecture 13: Generative ModelsUnsupervised Learning相比于从有标注的训练数据中学习f:x↦yf:x↦yf:x\mapsto y的有监督学习,无监督学习旨在学习无标注数据的隐含结构,包括聚类(K-means)、降维(PCA)、特征学习(Auto-encode)、密度估计等Generative Models: Give...
2018-05-07 21:50:53
185
原创 Lecture 13: Generative Models
CS231nLecture 13: Generative ModelsUnsupervised Learning相比于从有标注的训练数据中学习f:x↦yf:x↦yf:x\mapsto y的有监督学习,无监督学习旨在学习无标注数据的隐含结构,包括聚类(K-means)、降维(PCA)、特征学习(Auto-encode)、密度估计等Generative Models: Give...
2018-05-02 08:49:36
284
原创 Lecture 12: Visualizing and Understanding
CS231nLecture 12: Visualizing and UnderstandingVisualizeFilters of first layerfeature sof last layer: t-SNEVisualizing ActivationsOcclusion ExperimentsSaliency Maps: 梯度反传到输入Intermediate fe...
2018-04-25 16:37:06
187
原创 Lecture 11: Detection and Segmentation
CS231nLecture 11: Detection and SegmentationSemantic SegmentationLabel each pixel in the image with a category label Not differentiate instances, only care about pixels 最简单的想法:Sliding window,但...
2018-04-25 11:29:57
214
原创 Lecture 10: Recurrent Neural Networks
CS231nLecture 10: Recurrent Neural Networks RNN one many one Vanilla Image captioning many Sentiment Classification Machine Translation,Video classification on frame level...
2018-04-25 11:12:36
276
原创 Lecture 9: CNN Architectures
CS231nLecture 9: CNN ArchitecturesCase StudiesLeNet-5AlexNetZFNetVGGNetsmaller filters:Stack of three 3x3 conv (stride 1) layers has same effective receptive field as one 7x7 conv la...
2018-04-25 10:34:05
229
原创 Lecture 8: Deep Learning Software
CS231nLecture 8: Deep Learning SoftwareCPU < GPUCPU: Fewer cores, but each core is much faster and much more capable; great at sequential tasks GPU: More cores, but each core is much slower ...
2018-04-25 10:14:48
139
原创 Lecture 7: Training Neural Networks, Part 2
CS231nLecture 7: Training Neural Networks, Part 2OptimizationSGDw -= lr * gradLoss function has high condition number: ratio of largest to smallest singular value of the Hessian matrix is l...
2018-04-25 09:55:23
198
原创 Lecture 6: Training Neural Networks, Part I
CS231nLecture 6: Training Neural Networks, Part IReview回顾之前的内容,我们学习了神经网络的反向传播训练方法和CNN的结构,于是对于CNN我们可以用反向传播方法进行训练,具体方式是 1. 采样mini-batch 2. 前向传播获得loss 3. 根据loss进行反向传播梯度 4. 根据梯...
2018-04-25 09:29:37
247
原创 Lecture 5: Convolutional Neural Networks
CS231nLecture 5: Convolutional Neural NetworksHistoryRosenblatt Perceptron →→\to Rumelhart Back-propagation →→\to Hinton RBM →→\to DBN →→\to AlexNet Hubel & Wiesel →→\to FUKUSHIMA Neuroco...
2018-04-24 23:03:16
167
原创 Lecture 4: Backpropagation and Neural Networks
CS231nLecture 4: Backpropagation and Neural NetworksBackpropagation反向传播的理论基础:多元函数微分学中的链式法则 z=z(x,y),x=x(u,v),y=y(u,v)⇒dz=∂z∂xdx+∂z∂ydy=∂z∂x(∂x∂udu+∂x∂vdv)+∂z∂y(∂y∂udu+∂y∂vdv)⇒∂z∂u=∂z∂x∂x∂u+∂z...
2018-04-24 23:02:37
228
原创 Lecture 3: Loss Functions and Optimization
CS231nLecture 3: Loss Functions and Optimization上节课介绍了线性分类器,这节课介绍如何通过定义合适的loss函数(正如《统计学习方法》中介绍的统计学习三大要素之一的“策略”)并优化该loss得到合适的参数(“算法”)lossSVM loss/Hinge loss L(x,y)=∑i≠ymax(0,si−sy+1)L(x,y)=∑i...
2018-04-24 22:45:44
211
原创 Lecture2: Image Classification pipeline
CS231nLecture2: Image Classification pipeline图像分类计算机视觉的一项核心任务图像→→\to类别对于计算机而言,图像仅仅是一个数组/流形困难:视角变化、光照、变形、遮挡、背景混淆、类内差异没有显然的解决方案利用机器学习方法,从数据中学习NN:训练时间复杂度O(1)O(1)O(1),测试时间复杂度O(N)O(N)O(N)kNN:...
2018-04-24 21:53:51
216
原创 Lecture 1: Introduction
CS231nLecture 1: Introduction计算机视觉简史物种大爆炸时期生物开始有原始视觉→→\to文艺复兴时期Leonardo da Vinci的原始相机Obscura→→\to1959年Hubel和Wiesel关于视觉的研究→→\to1963年Larry Roberts利用Block对世界进行建模→→\to1966年MIT欲毕其功于一役→→\to1970s Dav...
2018-04-24 21:43:37
147
原创 组合数学引论
组合数学引论许胤龙、孙淑玲一、鸽巢原理Ramsey数习题二、排列组合加/减法原理、乘/除法原理排列从nnn元集合SSS中选出rrr 个元素将其按次序排列。其数目用ArnAnrA_n^r或P(n,r)P(n,r)P(n,r)表示。 Arn=n!(n−r)!Anr=n!(n−r)!A_n^r = \frac{n!}{(n-r)!}组合从nnn元集合SS...
2018-04-24 17:46:13
2106
原创 SPATIALLY TRANSFORMED ADVERSARIAL EXAMPLES
SPATIALLY TRANSFORMED ADVERSARIAL EXAMPLES本文发表在ICLR2018上引言传统的对抗样本生成方式都是加扰动,是一种像素值变换,本文提出一种空域变换生成对抗样本的方法stAdv,虽然基于此方法在传统的对抗样本生成评价指标中和原图像会有较大的LpLpL^p距离,但是从人的视觉感官上这种变换方式更真实,且更不容易被现有对抗攻击防御方法检测出来。 传...
2018-04-24 10:26:32
1623
空空如也
TA创建的收藏夹 TA关注的收藏夹
TA关注的人