CVPR人脸识别梳理

1:Feature Transfer Learning for Face Recognition with Under-Represented Data 

基于数据不足的数据集通过特征迁移学习人脸识别。

 

2:Disentangled Representation Learning for 3D Face Shape

将3D的人脸形状分解成认证部分(identity part)和表示部分(expression)

 

3: AdaptiveFace: Adaptive Margin and Sampling for Face Recognition     

自适应的margin softmax loss算法,用于处理人脸数据集不均衡问题

 

4:Low-Rank Laplacian-Uniform Mixed Model for Robust Face Recognition

解决人脸图像会因为遮挡,损坏等问题不能被正确识别的问题

 

5:Attribute-aware Face Aging with Wavelet-based Generative Adversarial Networks

用wavelet-based gan优化年轻人脸和年老人脸的匹配模糊问题。

 

6:Noise-Tolerant Paradigm for Training Face Recognition CNNs

大规模人脸数据集中难免会出现某些图片的标签标错了,标错了的数据集用于训练明显会有负面影响,但是如果查找数据集一张一张改正过来是非常耗时耗力的一件工作,本文解决这个问题。

 

7:Expressive Body Capture: 3D Hands, Face, and Body from a Single Image

单张人体的三维重建,里面又Pytorch源码,数据,(没找到代码)

 

8: Monocular Total Capture: Posing Face, Body, and Hands in the Wild

和7 功能上基本上差不多。

 

9:Boosting Local Shape Matching for Dense 3D Face Correspondence

两张人脸模型匹配

 

10:Combining 3D Morphable Models: A Large scale Face-and-Head Model

不懂

 

11:

AdaCos: Adaptively Scaling Cosine Logits for Effectively

Learning Deep Face Representations

提出一个loss   AdaCos,无超参自动训练优化网络,值得一读

 we propose a novel cosinebased softmax loss, AdaCos, which is hyperparameter-free and leverages an adaptive scale parameter to automatically strengthen the training supervisions during the train-

ing process. 

 

12:High-Quality Face Capture Using Anatomical Muscles

通过肌肉来获得高质量人脸信息。3D

 

13:FML: Face Model Learning from Videos

基于视频的3D人脸重建

 

14:

APDrawingGAN: Generating Artistic Portrait Drawings

from Face Photos with Hierarchical GANs

 

通过人脸生成绘画图像,多层次风格

 

15:Automatic Face Aging in Videos via Deep Reinforcement Learning

使用深度强化学习,在视频中检测人年龄。

 

16:

Multi-adversarial Discriminative Deep Domain Generalization for Face Presentation Attack Detection

有关于人脸方向攻击的

 

17:P2SGrad: Refined Gradients for Optimizing Deep Face Models。

解决超参调参问题

 

18:Semantic Component Decomposition for Face Attribute Manipulation

GAN相关,有用户编辑的人脸修正。

 

19:R3 Adversarial Network for Cross Model Face Recognition

本篇工作提出一个新方向,跨平台人脸识别?没太搞清楚。

 

20:Unsupervised Face Normalization with Extreme Pose and Expression in the Wild

无监督人脸归一化,统一同一人脸的差异性,使用gan

 

(21):3D Guided Fine-Grained Face Manipulation

利用三维信息细粒度微调人脸表情

 

(22):Joint Face Detection and Facial Motion Retargeting for Multiple Faces

一图多张人脸检测人脸关键点定位

 

(23):Hierarchical Cross-Modal Talking Face Generation

with Dynamic Pixel-Wise Loss

深度框架为级联gan,使用 Dynamic Pixel-Wise Loss跨模态说话人脸生成

 

(24):Unequal-training for Deep Face Recognition with Long-tailed Noisy Data

关于 long-tailed Noisy 数据集提出来的算法吧。

 

(25):Unequal-training for Deep Face Recognition with Long-tailed Noisy Data

视频流人脸的ADD(Automated deception detection)

 

(26):Learning to Regress 3D Face Shape and Expression from an Image without 3D Supervision

在没有2d到3D的数据集训练3D人脸形状回归和表示

 

(27):Efficient Decision-based Black-box Adversarial Attacks on Face Recognition

董胤蓬 人脸的攻击防御系统。

 

(28):FA-RPN: Floating Region Proposals for Face Detection

通过改进rpn的目标框进行人脸识别算法设计

 

(29):MMFace: A Multi-Metric Regression Network for Unconstrained Face Reconstruction

采用MM回归进行3D人脸重建,landmark

 

 

(29):Speech2Face: Learning the Face Behind a Voice

妈的,听到这个题目我就震惊了,通过声音来生成人脸,我怀疑这篇论文造假,但是我没有证据

 

 

(30): Led3D: A Lightweight and Efficient Deep Approach to Recognizing Low-quality 3D Faces

低质量人脸识别

 

(31):Face Parsing with RoI Tanh-Warping

人脸语法分析(应该是判断)

 

这里有关于旋转问题的解决方法,可以看看。其实在调研人脸方向的时候忽视了一个非常重要的问题就是人脸的倾斜角度,这个其实很关键,因为旋转过一定方向,人的识别就出现巨大偏差,2015时有论文试图解决这方面问题后面没有特地的跟踪。

 

(32):DSFD: Dual Shot Face Detector

 

人脸检测的多重突破。

 

(33):ArcFace: Additive Angular Margin Loss for Deep Face Recognition

主要提出Additive Angular Margin Loss,对triplet loss改进,可读

 

(34):Deep Tree Learning for Zero-shot Face Anti-Spoofing

关于人脸反欺骗的深度树学习,可读

 

(35):Self-Supervised Adaptation of High-Fidelity Face Models for

Monocular Performance Tracking

二维单张人脸的检测和三维重建,摆正角度。

 

 

(36):Decorrelated Adversarial Learning for Age-Invariant Face Recognition

关于不同年龄的人脸检测,使用GAN。

 

(37):Face Anti-Spoofing: Model Matters, So Does Data

人脸反欺骗,这个有点意思

 

还是采用人打印图片来测试,可读。

 

(38):Group Sampling for Scale Invariant Face Detection

这篇论文发现(multi layer prediction is not necessary. Faces at all scales can be well detected with features from a single layer of the network.) 多层预测不是必要的,单层网络也可以提取的很好

 

(39):UniformFace: Learning Deep Equidistributed Representation for Face Recognition

关于特征识别,2分支3分支都在强调增大类间差距,减小类内差距,但是这样做忽视了人脸特征在整体特征的分布情况,本文提出新算法。

 

(40):Learning to Cluster Faces on an Affinity Graph

人脸识别需要大量的数据集,但是很多情况下不满足,所以需要用聚类这种无监督方法来改进。

 

(41):GANFIT: Generative Adversarial Network Fitting for High Fidelity 3D Face Reconstruction

2维人脸的三维重建,但是在重建过程中可以保持很高的清晰度。

 

(42):RegularFace: Deep Face Recognition via Exclusive Regularization

所有论文中最值得一读的论文,代码量很少,原理简单,可以复用。

 

(43):Towards High-fidelity Nonlinear 3D Face Morphable Model

单张人脸的3D重建,高分辨率。

 

(44):Linkage Based Face Clustering via Graph Convolution Network

采用GCN(图卷积网络)做聚类。

 

(45):Dense 3D Face Decoding over 2500FPS: Joint Texture & Shape Convolutional Mesh Decoders

不想懂。

 

 

(46)MVF-Net: Multi-View 3D Face Morphable Model Regression

多视图3D人脸回归

 

(47):A Dataset and Benchmark for Large-scale Multi-modal Face Anti-spoofing

关于多尺度多模型人脸反欺骗,数据集和衡量标准。

 

你可能感兴趣的:(CVPR人脸识别梳理)