Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

videogan资料 #4

Open
2 tasks done
zdx3578 opened this issue Oct 17, 2017 · 3 comments
Open
2 tasks done

videogan资料 #4

zdx3578 opened this issue Oct 17, 2017 · 3 comments
Milestone

Comments

@zdx3578
Copy link
Member

zdx3578 commented Oct 17, 2017

gan video 两周
第一周, 熟悉已有论文,选用一个算法复现。使用自己的数据集,调参。
第二周,分析隐变量的语义相关信息,自动驾驶的转向角度和z的关系 。

@waxz waxz added this to the 10月计划 milestone Oct 17, 2017
@zdx3578
Copy link
Member Author

zdx3578 commented Oct 18, 2017

ref:
VAE:
Adversarial Variational Bayes: Unifying VAE and GAN
beta vae

GAN:

https://scholar.google.com/scholar?um=1&ie=UTF-8&lr&cites=7977179880967579431
https://scholar.google.com/scholar?um=1&ie=UTF-8&lr&cites=12629733064507558057

@waxz
Copy link
Contributor

waxz commented Oct 18, 2017

note
1.Temporal Generative Adversarial Nets with Singular Value Clipping

  • 从随机采样的z0生成整个视频序列
  • conditional tgan,根据视频类别标签和z0生成成视频

2.Generating Videos with Scene Dynamics

  • 从随机采样的z0生成整个视频序列
  • 引入了静态的背景和动态的前景的先验知识,用于对物体动作跟踪问题的建模

3.Learning to Generate Time-Lapse Videos Using Multi-Stage Dynamic Generative Adversarial Networks

  • 给定一张静态图像,生成视频序列
  • 两个Gan叠加,第一个Base Gan生成粗糙/模糊的图像,第二个REfine Gan在第一个的基础上进行精细化
  • Base Gan,输入的是初始图像复制叠加形成的图像序列X,输出图像序列Y1。生成器采用Unet结构,损失函数是判别器的loss和L1 loss
  • Refine net输入Base Gan输出的图像序列Y1,输出图像序列Y2。损失函数为判别器的loss和L1 loss,加上Gram matrix loss ,根据Y,Y1,Y2在判别器某一层输出计算的Gram matrix。采用Gram matrix loss的目的是避免Y2只是简单的逼近Y1,因为Y1已经足够真实。
  1. MoCoGAN: Decomposing Motion and Content for Video Generation
  • 引入动作和内容先验知识,将噪声Z分为Zc内容噪声和Zm动作噪声
  • 利用RNN从动作空间获得Zm分布Rm,随机变量e --> RNN --> Zm
  • 通过锁定Zm或Zc改变动作或环境
  • 整体模型含4个网络,分别为:
    • 获取动作分布 - GRU
    • 生成器Gi 、 判别器Di - DCGAN - 控制图像生成质量
    • 判别器Dv - spatio-temporal CNN 控制视频&动作生成质
  • Conditional MoCoGAN, 将label与E一同输入RNN

@fzd9752
Copy link
Contributor

fzd9752 commented Oct 18, 2017

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants