search

AntixK / PyTorch-VAE

A Collection of Variational Autoencoders (VAE) in PyTorch.
Apache License 2.0
6.38k stars 1.05k forks source link

Access the latent space #64

Open gfiameni opened 2 years ago

gfiameni commented 2 years ago

Hi all, I am training a Vanilla VAE on a custom data set and interested in projecting the constructed latent space. Any suggestion on how to do so?

Thanks!

jprachir commented 2 years ago

Hi, have you figured it out? I am also interested in learning how much time it takes to draw samples from a pre-trained model.

KfirdausMT commented 1 year ago

Hi.. Have you both sort this out. Need help on this too

ranabanik commented 5 months ago

If your data is mostly multivariate Gaussian, you can use Gaussian Mixture Models to classify the data and maybe project on 2D, 3D space. This has already existing applications. If your data is not multivariate Gaussian, then the latent space can be projected and classified in some tSNE or non-linear clustering method. This is my hypothesis...