![](https://img-blog.csdnimg.cn/20201014180756922.png?x-oss-process=image/resize,m_fixed,h_64,w_64)
读书笔记
文章平均质量分 55
o0Helloworld0o
怕是要翻水水哦
展开
-
Learning Signed Distance Field for Multi-view Surface Reconstruction(ICCV21)
暂无原创 2022-07-14 10:02:17 · 226 阅读 · 1 评论 -
Unsupervised Learning of Probably Symmetric Deformable 3D Objects from Images in the Wild(CVPR20)
本文的效果其实也一般,去网站demo跑一下就知道了,一个明显的瑕疵是眼睛容易被预测成尖的;尽管如此,还是可以从源代码中学习到很多东西(因为支持透视投影)同时,可以对比一下DECA,因为DECA也是使用了displacement map本文的方法对于嘟嘴,也无法重建出来,一是因为嘟嘴被投影成图像后,信息丢失太多了,难度很大;二是数据集中本身嘟嘴的图像就不多3. Method本文的方法不仅局限于人脸,只要是同一个类别的object就行As we have only raw images to lea原创 2021-07-09 19:17:03 · 190 阅读 · 0 评论 -
Face Alignment Across Large Poses: A 3D Solution(CVPR16,TPAMI17)
Face Alignment Across Large Poses: A 3D Solution(CVPR16)AbstractFace alignment, which fits a face model to an image and extracts the semantic meanings of facial pixels, has been an important topic in CV community.这算是对Face Alignment的含义的权威解释吗yaw=0~45°属原创 2021-04-03 16:02:17 · 403 阅读 · 0 评论 -
img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation(CVPR21)
可视化pose_references/vertices_trans.npyc=0, [-0.891652, 0.890319], span=1.781972c=1, [-0.975868, 1.000126], span=1.975995c=2, [-0.751428, 0.774013], span=1.525441center = [-0.00005079 -0.00001977 -0.00001119]原创 2021-02-28 10:53:01 · 1569 阅读 · 2 评论 -
FaceScape:a Large-scale High Quality 3D Face Dataset and Detailed Riggable 3D Face Prediction(CVPR)
源代码理解固定3D关键点,2D投影关键点,求s, R, tdef _optimize_rigid_pos(self, scale, trans, rot_vector, lm_pos_3D, lm_pos)scale: 标量trans: 2维向量rot_vector: 3维向量lm_pos_3D: (68, 3)lm_pos: (68, 2)核心调用from scipy.optimize import least_squaresresult = least_squares(self._原创 2021-02-19 09:09:48 · 471 阅读 · 0 评论 -
Towards Fast, Accurate and Stable 3D Dense Face Alignment(ECCV20)
从源代码理解3DDFA_V2的推理过程默认crop_policy = 'box',bbox宽和高的平均值记为old_size,找到bbox的中心(center_x, center_x),从中心向四周扩展尺寸为int(old_size * 1.58),从而截取出一个正方形resize截取正方形,使得尺寸为120x120,输入网络,输出为一个62维向量解析62维向量...原创 2021-02-15 18:26:48 · 1030 阅读 · 2 评论 -
Rotate-and-Render: Unsupervised Photorealistic Face Rotation from Single-View Images(CVPR20)
首先进行一些科普3D Face FittingV=[v1,v2,⋯ ,vn]∈Rn×3\mathbf{V}=\left [ v_1, v_2, \cdots, v_n \right ]\in\mathbb{R}^{n\times3}V=[v1,v2,⋯,vn]∈Rn×3表示一个含有nnn个顶点的3D mesh将3D mesh按照P={f,R,h2d}\mathbf{P}=\left \{ f, \mathbf{R}, \mathbf{h}_\text{2d} \right \}P={f,R,h2原创 2020-11-19 09:51:37 · 451 阅读 · 0 评论 -
StyleRig: Rigging StyleGAN for 3D Control over Portrait Images(CVPR20 oral)
4. Semantic Rig Parameters原创 2020-11-17 15:58:53 · 644 阅读 · 3 评论 -
MobileNetV2: Inverted Residuals and Linear Bottlenecks
PyTorch代码原创 2020-10-23 11:11:57 · 152 阅读 · 0 评论 -
MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
定义输入feature map尺寸为DF×DF×MD_F\times D_F\times MDF×DF×M,输出feature map尺寸为DF×DF×ND_F\times D_F\times NDF×DF×N,假设卷积前后空间维度不变,通道数由MMM变为NNN原创 2020-10-21 16:00:54 · 131 阅读 · 0 评论 -
Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization(ICCV17)
Perceptual Losses for Real-Time Style Transfer and Super-Resolution(ECCV16)给定输入图像xxx,经过一个网络得到yyy,同时有content image ccc和style image sss,使用一个VGG19来计算loss,令yyy的content与ccc相似,同时令yyy的style与sss相似...原创 2020-09-09 19:54:39 · 368 阅读 · 0 评论 -
Image Style Transfer Using Convolutional Neural Networks(CVPR16)
Abstract之前的工作不太成功,是因为缺乏一种表示图像semantic information的representations,用来分离图像的content和style1. IntroductionTransferring the style from one image onto another can be considered a problem of texture transfer.style transfer本质上是texture transfer,所以本文的目标是按照source原创 2020-09-09 15:19:52 · 767 阅读 · 0 评论 -
Reusing Discriminators for Encoding: Towards Unsupervised Image-to-Image Translation(CVPR20)
No Independent Component for Encoding (NICE).原创 2020-08-31 15:35:19 · 649 阅读 · 0 评论 -
开源人脸数据集
VGGFace2test_list.txt 共169396行testn000001n009294原创 2020-07-03 20:22:39 · 447 阅读 · 0 评论 -
MarioNETte: Few-shot Face Reenactment Preserving Identity of Unseen Targets(AAAI20)
MarioNETte ArchitectureFig.2展示了MarioNETte的框架图给定driver image x\mathbf{x}x,一组target images {yi}i=1⋯K\left \{ \mathbf{y}^i \right \}_{i=1\cdots K}{yi}i=1⋯K,整个framework输出一幅Reenacted image注意:driver x\...原创 2020-03-05 20:21:11 · 702 阅读 · 0 评论 -
CSGAN: Cyclic-Synthesized Generative Adversarial Networks for Image-to-Image Transformation
II. PROPOSED CSGAN ARCHITECTURE数据集X∈{(Ai),(Bi)}i=1nX\in\left \{ (A_i), (B_i) \right \}_{i=1}^nX∈{(Ai),(Bi)}i=1n,包含nnn个样本,每个样本包含来自domain AAA和BBB的2幅paired images学习目标是2个生成器:GAB:A→BG_{AB}: A\rightarr...原创 2020-03-05 19:12:01 · 445 阅读 · 0 评论 -
Landmark Assisted CycleGAN for Cartoon Face Generation
3. Our Method3.1. Review of CycleGAN给定来自两个domain的unpaired training samples x∈X,y∈Yx\in X, y\in Yx∈X,y∈Y,对于其从XXX到YYY的mapping GX→YG_{X\rightarrow Y}GX→Y,及其判别器DYD_YDY,adversarial loss定义如下LGAN(GX→Y,D...原创 2020-02-20 10:38:54 · 1196 阅读 · 0 评论 -
Make a Face: Towards Arbitrary High Fidelity Face Manipulation(ICCV19)
3. Method定义face image x∈Xx\in Xx∈X,给定target facial structural information ccc,学习一个mapping G\mathcal{G}G,将xxx转换为output image x~\tilde{x}x~原创 2020-02-10 16:25:19 · 397 阅读 · 0 评论 -
Face Video Generation from a Single Image and Landmarks
3. Proposed Framework本文提出MotionGAN,给定source image sss及其landmark lll,还有一段target landmark序列 l1T=[l1,l2,⋯ ,lT]l_1^T=\left [ l_1, l_2, \cdots, l_T \right ]l1T=[l1,l2,⋯,lT],生成的一段video f~1T=[f~1,f~2,⋯ ...原创 2020-02-06 11:48:08 · 535 阅读 · 0 评论 -
Few-Shot Adversarial Learning of Realistic Neural Talking Head Models(ICCV19)
3.2. Meta-learning stagesimulating episodes of K-shot learning (K = 8 in our experiments)随机选取第iii个视频xi\textbf{x}_ixi中的第ttt帧xi(t)\textbf{x}_i(t)xi(t),接着再从这个视频中额外抽取KKK帧,也就是KKK个index,记为s1,s2,⋯ ,sKs...原创 2020-02-03 16:46:33 · 616 阅读 · 0 评论 -
【Note】pytorch-CycleGAN-and-pix2pix
下载数据集summer2winter_yosemite,文件夹结构如下summer2winter_yosemite ├─ testA 310幅256x256图像 ├─ testB 239幅 ├─ trainA 1232幅 └─ trainB 963幅训练模型python train.py --dataroot datasets/summer2winter_yosemite ...原创 2020-01-20 15:11:03 · 1739 阅读 · 0 评论 -
Semantic Image Synthesis with Spatially-Adaptive Normalization(CVPR19)
3. Semantic Image Synthesis定义m∈LH×W\mathbf{m}\in\mathbb{L}^{H\times W}m∈LH×W为semantic segmentation mask,其中L\mathbb{L}L是一系列整数用于指定semantic labelSpatially-adaptive denormalization定义hi∈RN×Ci×Hi×Wi\math...原创 2020-01-17 15:08:52 · 701 阅读 · 0 评论 -
FSGAN: Subject Agnostic Face Swapping and Reenactment(ICCV19)
3. Face swapping GAN定义source face image为IsI_sIs,target face image为ItI_tItFSGAN包含3个部分原创 2020-01-16 16:41:20 · 594 阅读 · 0 评论 -
Learning Continuous Face Age Progression: A Pyramid of GANs(CVPR18扩展)
1 INTRODUCTION本文是CVPR18的扩展3 METHOD3.1 Overviewloss包括the traditional squared Euclidean loss、the GAN loss、the identity loss结构上,判别器是pyramid-structured discriminator3.2 Generator生成器是Encoder-Decode...原创 2020-01-15 14:36:54 · 494 阅读 · 0 评论 -
Image-to-Image Translation with Conditional Adversarial Networks(CVPR17)
本文提供的代码https://github.com/phillipi/pix2pix,因此本文的方法被称为pix2pix1. Introductionimage-to-image translation的定义:translating one possible representation of a scene into another, given sufficient training da...原创 2020-01-10 10:59:24 · 136 阅读 · 0 评论 -
Age Progression and Regression with Spatial Attention Modules(AAAI20)
MethodProblem Formulation定义young face image为Iy\mathbf{I}_yIy,对应的age为αy\bm{\alpha}_yαy给定目标age αo\bm{\alpha}_oαo(要求αo>αy\bm{\alpha}_o\gt\bm{\alpha}_yαo>αy),我们希望学习一个age progressor GpG_pGp,...原创 2020-01-09 10:50:42 · 630 阅读 · 0 评论 -
FaceShifter: Towards High Fidelity And Occlusion Aware Face Swapping
3. Methods定义XsX_sXs为source image,提供identity信息,XtX_tXt为target image,提供attribute信息(包括pose、expression、scene lighting和background)FaceShifter包含2个stage,在第1个stage中,采用Adaptive Embedding Integration Networ...原创 2020-01-08 11:13:50 · 1172 阅读 · 0 评论 -
One-shot Face Reenactment(BMVC19)
3 Approach给定source face xsx_sxs,包含了pose guidance,以及target face xtx_txt,包含了reference appearance,学习的目标是生成一幅图像包含xsx_sxs的pose/expression,以及xtx_txt的identity...原创 2020-01-02 16:02:19 · 1069 阅读 · 0 评论 -
A Style-Based Generator Architecture for Generative Adversarial Networks(CVPR19)
2. Style-based generator如Figure 1a所示,传统的生成器的输入层负责接收一个latent code z∈Zz\in\mathcal{Z}z∈Z(其实就是GAN中的noise,作者为了和生成器中的noise区分,此处使用术语latent code)用于生成图像如Figure 1b所示,本文的抛弃了传统的生成器设计,令生成器的输入层负责接收一个learned cons...原创 2019-12-26 16:51:38 · 485 阅读 · 0 评论 -
StarGAN v2: Diverse Image Synthesis for Multiple Domains
输入图像x∈Xx\in\mathcal{X}x∈X,arbitrary domain y∈Yy\in\mathcal{Y}y∈Y原创 2019-12-21 11:22:39 · 1155 阅读 · 1 评论 -
StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation(CVPR18)
3. Star Generative Adversarial Networks3.1. MultiDomain ImagetoImage Translation学习目标是训练一个能够在multiple domains之间相互生成的生成器GGG定义xxx为输入图像,yyy为生成图像,ccc为target domain label,于是有KaTeX parse error: Undefined ...原创 2019-12-20 11:56:34 · 326 阅读 · 0 评论 -
LOGAN:Latent Optimisation for Generative Adversarial Networks
GAN的min-max优化目标可以抽象为minθD maxθG Ex∼p(x)[hD(x)]+Ez∼p(z)[hG(D(G(z)))](1)\underset{\theta_D}{\min}\ \underset{\theta_G}{\max}\ \mathbb{E}_{x\sim p(x)}\left [ h_D(x) \right ]+\mathbb{E}_{z\s...原创 2019-12-18 19:47:56 · 498 阅读 · 0 评论 -
AttGAN: Facial Attribute Editing by Only Changing What You Want(TIP19)
III. ATTRIBUTE GAN (ATTGAN)前提:所有attribute都是binary型的A. Testing Formulation定义输入图像为xa\mathbf{x^a}xa,包含nnn个attribute a=[a1,⋯ ,an]\mathbf{a}=\left [ a_1, \cdots, a_n \right ]a=[a1,⋯,an]encoder网络Genc...原创 2019-12-16 16:55:41 · 510 阅读 · 0 评论 -
Video-to-Video Synthesis(NeurIPS18)
image-to-image translation是一个被广泛研究的问题,而video-to-video synthesis则是它的升级版,受到的关注较少如果不考虑temporal dynamics,直接使用image-to-image translation的方法会生成不连贯(incoherent)、低质量的视频1 Introduction据作者所知,之前还没有工作专门提出a gener...原创 2019-12-16 14:45:05 · 3303 阅读 · 0 评论 -
STGAN: A Unified Selective Transfer Network for Arbitrary Image Attribute Editing(CVPR19)
3. Proposed Method3.1 Limitation of Skip Connections in AttGANUnfortunately, downsampling irreversibly diminishes spatial resolution and fine details of feature map, which cannot be completely reco...原创 2019-12-12 14:37:56 · 526 阅读 · 0 评论 -
GANimation: Anatomically-aware Facial Animation from a Single Image(ECCV18)
3 Problem Formulation定义输入图像Iyr∈RH×W×3\mathbf{I}_{\mathbf{y}_r}\in\mathbb{R}^{H\times W\times3}Iyr∈RH×W×3,yr=(y1,⋯ ,yN)T\mathbf{y}_r=\left ( y_1,\cdots,y_N \right )^Tyr=(y1,⋯,yN)T表示NNN个Action Uni...原创 2019-12-10 21:20:04 · 440 阅读 · 0 评论 -
Towards Open-Set Identity Preserving Face Synthesis(CVPR18)
3. Identity Preserving GANs输入两幅图像(xs,xa)(x^s, x^a)(xs,xa),xsx^sxs指定identity信息,xax^axa指定attribute信息(包括pose, emotion, illumination, and even background)作者描述得比较保守,并没有保证保留background信息生成一幅图像x′x'x′,拥有xsx...原创 2019-12-10 11:16:28 · 456 阅读 · 1 评论 -
Bag of Tricks for Image Classification with Convolutional Neural Networks
目前性能最好的单模型为NASNet-A,在ISLVRC2012验证集上的Top1Acc为82.7%使用文中提到的Trick来训练ResNet-50,比ResNet后续的变体中的结果都要高原创 2018-12-20 16:12:15 · 134 阅读 · 2 评论 -
FAST AI Deep Learning Note
https://www.bilibili.com/video/av18904696/?p=890:21使用dict创建DataFrame时,不保证列的顺序,再指定columns就可以保证列的顺序91:50在Image Classification中,通常目标物体很大,且位于图像中心位置,可以做RandomCrop而在Object Detection中,存在很多小物体,并且小物体可能位于图...原创 2018-09-22 11:12:27 · 387 阅读 · 0 评论 -
Fast AI ML Note
Lesson1,https://www.bilibili.com/video/av2335658008:17在notebook开始处写下如下代码%load_ext autoreload%autoreload 2%matplotlib inline能够为后续的restart提供遍历(暂时体会不到)...原创 2018-09-25 15:29:14 · 801 阅读 · 0 评论