How to use core methods in my own latent space diffusion model?

[fushen1228]

how to use the image y to be restored and the mask in the latent space? Can I use the VAE encoder to encode them into the latent space? Then, can I decode the final restored x back to pixel space using the decoder?

[wyhuai]

The encoder is not a linear operator, so Range Null-space Decomposition may hold in that case.