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

wyhuai / DDNM

[ICLR 2023 Oral] Zero-Shot Image Restoration Using Denoising Diffusion Null-Space Model

MIT License

1.16k stars 83 forks source link

How to use core methods in my own latent space diffusion model? #78

Open fushen1228 opened 1 month ago

fushen1228 commented 1 month ago

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 commented 1 month ago

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