pytorch-sample-codes
pytorch-sample-codes copied to clipboard
A question about sampling in the testing phase
Hi, I benefited a lot from your blog about CVAE, but I have a question about sampling in the testing phase.
During testing, you sample a latent variable z from a normal distribution. But I wonder why don't sample from the posterior distribution, because after training, it can truly reflect the distribution of the samples and is near to the normal distribution.