microsoft / i-Code

MIT License
1.65k stars 162 forks source link

VAE in Training of each LDM in CoDi #131

Open NakataKoo opened 1 month ago

NakataKoo commented 1 month ago

Thank you for the great project.

Am I right in thinking that "in learning each Latent Diffusion Model in CoDi, the Encoder-Decoder part of VAE uses a pre-trained model, so only the U-Net part is the target of training"?

Also, do you have any plans to release the study scripts?

danielemolino commented 1 month ago

I'll to know it too