VAE reconstruction loss does not get calculated right

Question

I wanted to implement a variational autoencoder and train it on my texture dataset which consists of colored images of the size 512x512.

For the VAE implementation, I