Confusion about Two-Phase Encoding
ChrisWang13 opened this issue · comments
Chris commented
I'm new to diffusion models and trying to grasp the purpose behind the two-phase encoding process in ddpm.py
.
Could someone explain why there's need to have encode_first_stage
and encoder_posterior
encoding steps for the input image (batch['image']
)?
Why the need for two encoding phases? Clear insights into this design decision would greatly aid my understanding.
Thanks for your help!