search

sail-sg / MDT

Masked Diffusion Transformer is the SOTA for image synthesis. (ICCV 2023)
Apache License 2.0
500 stars 35 forks source link

Do I need to train a VAE myself? #40

Closed feifeitu1 closed 4 months ago

feifeitu1 commented 4 months ago

Very good job!! I am a newbie who is still in the exploration stage of deep learning. And I want to know if your training process already include transforming images into latent dimensions?Do I need to pre-train a VAE myself? Looking forward to your answer very much. Thank you very much!!

gasvn commented 4 months ago

You can use a pretrained VAE, such as the VAE from stable diffusion

feifeitu1 commented 4 months ago

Oh，Thank you for your reply. Does the training process of MDT directly load the pre-trained VAE?

gasvn commented 4 months ago

Yes, we follow this setting from DiT.

feifeitu1 commented 4 months ago

OK, Thank you!!