"Generative Models"의 두 판 사이의 차이
ph
(→VAE) |
|||
| 7번째 줄: | 7번째 줄: | ||
=VAE= | =VAE= | ||
* http://kvfrans.com/variational-autoencoders-explained/ | * http://kvfrans.com/variational-autoencoders-explained/ | ||
| − | * autoencoder와 동일하나 latent vector를 생성할 때 (unit) gaussian으로만 생성하도록 constraint를 줌. 그래서 unit gaussian random variable로부터 generate. | + | ** autoencoder와 동일하나 latent vector를 생성할 때 (unit) gaussian으로만 생성하도록 constraint를 줌. 그래서 unit gaussian random variable로부터 generate. |
| − | * In practice, there's a tradeoff between how accurate our network can be and how close its latent variables can match the unit gaussian distribution. | + | ** In practice, there's a tradeoff between how accurate our network can be and how close its latent variables can match the unit gaussian distribution. |
| − | * latent vector를 바로 만들지도 않고 mean, std만 만들어낸다. | + | ** latent vector를 바로 만들지도 않고 mean, std만 만들어낸다. |
| − | * we can compare generated images directly to the originals, which is '''''not possible''''' when using a GAN. | + | ** we can compare generated images directly to the originals, which is '''''not possible''''' when using a GAN. |
* [https://jmetzen.github.io/2015-11-27/vae.html VAE in tensorflow] | * [https://jmetzen.github.io/2015-11-27/vae.html VAE in tensorflow] | ||
= GAN, VAE, pixel-rnn (by OpenAI)= | = GAN, VAE, pixel-rnn (by OpenAI)= | ||
https://blog.openai.com/generative-models/ | https://blog.openai.com/generative-models/ | ||
2017년 5월 4일 (목) 14:20 판
GAN
- http://kvfrans.com/generative-adversial-networks-explained/
- How to Train a GAN? Tips and tricks to make GANs work
- Generative Adversarial Networks (GANs) in 50 lines of code (PyTorch)
- real/fake 두 데이터 셋이 있어야 한다. D를 먼저 훈련시키고 다음에 G를 훈련시킨다.
VAE
- http://kvfrans.com/variational-autoencoders-explained/
- autoencoder와 동일하나 latent vector를 생성할 때 (unit) gaussian으로만 생성하도록 constraint를 줌. 그래서 unit gaussian random variable로부터 generate.
- In practice, there's a tradeoff between how accurate our network can be and how close its latent variables can match the unit gaussian distribution.
- latent vector를 바로 만들지도 않고 mean, std만 만들어낸다.
- we can compare generated images directly to the originals, which is not possible when using a GAN.
- VAE in tensorflow