search

zoubohao / DenoisingDiffusionProbabilityModel-ddpm-

This may be the simplest implement of DDPM. You can directly run Main.py to train the UNet on CIFAR-10 dataset and see the amazing process of denoising.
MIT License
1.48k stars 156 forks source link

Hello, I would like to ask what is the meaning of posterior_var #3

Closed zhufq00 closed 1 year ago

zhufq00 commented 1 year ago

I wonder why is it written in 76 lines instead of 77 lines in the picture? It doesn't look like a big difference 企业微信截图_16655574385306

zhufq00 commented 1 year ago

企业微信截图_1665557775712 I can understand the use of self.posterior_var or self.betas to represent var but I don't understand why self.posterior_var and self.betas are mixed together to represent var

zoubohao commented 1 year ago

Sorry, I forgot why I used this code... But I tried to replace it with  self. betas, It seems that there is no significant change in producing the image at the same quality level. You can treat it with betas.

zhufq00 commented 1 year ago

hhhh,ok,thank you very much