1konny/vq-vae

Pytorch Implementation of "Neural Discrete Representation Learning"

Neural Discrete Representation Learning, VQ-VAE

Pytorch implementation of Neural Discrete Representation Learning

Requirements

• python 3.6
• pytorch 0.2.0_4
• visdom
  
  

RESULT : MNIST

RESULT : CIFAR10

reconstruction of randomly selected, fixed images

reconstruction of random samples

you can reproduce similar results by :

python main.py --dataset CIFAR10 --batch_size 100 --k_dim 256 --z_dim 256

To do:

• visdom -> tensorboardX
• learning prior p(z) using PixelCNN
• image sampling( dummy input => (PixelCNN) => Z_dec => (Decoder) => image )
• add references and acknowledgements

vq-vae's Issues

Why it can generate different images?

By setting K=256, it seems that we can only generate 256 different images? Since there are only 256 different discrete embedding vectors in the latent space. Why it can generate different images?

Hi, I want to train VQ-VAE on my own data set. Could you please help me for that!!

small question about the vq-vae paper

hello!
thank you for your great work!

I have a question about loss function in paper.
L = log p(x|z(x)) + ||sg[z(x)] − e|| + β||z(x) − sg[e]||

the author mentioned that a third term exists because e can grow arbitrarily if it doesn't train as fast as the encoder parameters.
but I see that term only helps the encoder to be trained faster.

will it help the e to be trained faster too? but I assume that sg[e] is meaning that the e won't be trained by the term.
I hope this isn't a silly question ;) thx in advance.