1
$\begingroup$

According to the litterature and specifically to this paper, the wasserstein autoencoders is an encoder-decoder architecture. So it must contain encoder and decoder parts.

in the algorithm description of GAN-based WAN, in the requirements, there were initialization of encoder, decoder and latent discriminator.

In parallel, I saw many implementations of GAN based WAE on github in tensorflow, there were all implemented through a discrminator and generator parts. No autoencoder architecture. So my question, are those implementations correspond to another architecture beside the WAE-GAN ? or maybe I didn't understand the connection between the paper and the implementation.

Here you will find some links for implementation:

implementation 1 implementation 2 implementation 3

$\endgroup$
1
$\begingroup$

Actually, There is Wasserstein-GAN which is a GAN network but uses the wasserstein distance to calculate the loss. So, it has the discriminator and generator networks. The implementation 2 and implementation 3 correspond to that network.

However, my target was the Wasserstein autoencoders which has encoder, decoder and discriminator parts. The implementation 1 corresponds to Wasserstein autoencoders and not to Wasserstein GAN.

It was confusing for me.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.