VAE reconstruction model

Question

I am doing a Variational Autoencoder (VAE). Each input is of shape [64, 3, 256, 256].

Here is the network model :

class VAE(nn.Module):