Paper Reading

【ICCV19】
Face De-occlusion using 3D Morphable Model and Generative Adversarial Network
使用3DMM生成一幅像是换脸一样的图像,然后与原图拼接,送入GAN生成去遮挡图像,由于是有ground-truth的,所以自带一项MAE,除此之外,判别器分为global和local,总体来说,没什么新意

【AAAI20】
Facial Action Unit Intensity Estimation via Semantic Correspondence Learning
with Dynamic Graph Convolution

【CVPR20】
Your Local GAN

Cascade EF-GAN: Progressive Facial Expression Editing with Local Focuses
专做人脸表情编辑,对标GANimation
Framework层面是不是一步到位直接转换到target expression,而是生成一些中间的过渡,成为级联,有一定新意
网络结构层面则是Global+Local的方式,新意不大

Controllable Person Image Synthesis with Attribute-Decomposed GAN
任务是,给定source image(模特穿衣照),指定一个新的人体pose关键点,生成该模特改变人体pose后的图像,详见链接

Neural Head Reenactment with Latent Pose Descriptors
三星团队继续探索Neural Head Reenactment,该团队上一篇文章是Few-Shot Adversarial Learning of Realistic Neural Talking Head Models(ICCV19)

MaskGAN: Towards Diverse and Interactive Facial Image Manipulation
与SPADE有渊源,但是把问题缩小到了由semantic segmentation mask生成真实人脸,方法读起来有点费劲

Disentangled and Controllable Face Image Generation via 3D Imitative-Contrastive Learning
任务是,指定5个Disentangled人脸factor:identity、expression、illumination、pose、others,生成真实人脸图像
方法中使用的各种trick其实都已经很常见了

StyleRig: Rigging StyleGAN for 3D Control over Portrait Images
3D人脸建模中的3个factor:face pose、expression、scene illumination
At test time, our method generates images of faces with the photorealism of StyleGAN, while providing explicit control over a set of semantic control parameters.
简而言之,就是指定semantic control parameters,生成真实人脸(生成效果媲美StyleGAN)

Interpreting the Latent Space of GANs for Semantic Face Editing

(src_img, ref_img)模式的GAN,FaceShifter、StarGAN2、PSGAN

【arxiv】
Manipulated Face Detector: Joint Spatial and Frequency Domain Attention Network
一种结合空域频域信息,以及注意力机制的方法,有效检测编辑人脸

What comprises a good talking-head video generation?: A Survey and Benchmark
人脸驱动的综述

Kunster - AR Art Video Maker - Real time video neural style transfer on mobile devices
移动端实时风格迁移,初步阅读,值得总结

ICE-GAN: Identity-aware and Capsule-Enhanced GAN for Micro-Expression Recognition and Synthesis
微表情识别

DeepFaceLab: A simple, flexible and extensible face swapping framework
换脸技术

High Resolution Face Age Editing
年龄编辑,似乎github上的编辑效果不明显

The Creation and Detection of Deepfakes: A Survey
换脸检测综述

CONFIG: Controllable Neural Face Image Generation
研究人脸各种attribute如何生成

Regularization Methods for Generative Adversarial Networks: An Overview of Recent Studies
GAN中正则化方法的综述

FReeNet: Multi-Identity Face Reenactment
landmark换脸

你可能感兴趣的:(读书笔记)