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

Mask Fill #7

Open MaxxP0 opened 1 year ago

MaxxP0 commented 1 year ago

Can you maybe provide code on how to do Inpainting as seen in the paper?

gasvn commented 1 year ago

Okay, I will clean the code and release it.

MaxxP0 commented 1 year ago

Thank you very much because i struggle to understand as the model seems to predict the noise and not x directly

MaxxP0 commented 1 year ago

i guess the question is does the model predict the noise or the sample? because in the paper it sounds like it predicts the sample and in the code it seems to predict the noise. I would think that the model predicts the sample because that would allow it to do inpainting as described in the paper. i would really appreciate if anyone can clear up my confusion.