视频编码论文
Posted hhhhhxh
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了视频编码论文相关的知识，希望对你有一定的参考价值。
Deep Generative Video Compression(NIPS2019)
Based on VAE
Decoder(use stochastic recurrent variational auto-encoder): (z_t=)latent variables, (f=)global variables(optionally)
[p_ heta(x_{1:T},z_{1:T},f)=p_ heta(f)p_ heta(z_{1:T})prod_{t=1}^{T}p_ heta(x_t|z_t,f) p_ heta(x_t|z_t,f)=Laplace(mu_ heta(z_t,f),lambda^{-1}1) \widetilde{x}_t=mu_ heta(z_t,f)=decoder;mean
]
Encoder(use amortized variational inference): (f=)global variables inferred from all video frames, (z_t) inferred from a single frame (x_t)
[q_phi(z_{1:T},f|x_{1:T})=q_phi(f|x_{1:T})prod_{t=1}^{T}q_phi(z_t|x_t)
]
以上是关于视频编码论文的主要内容，如果未能解决你的问题，请参考以下文章