search

cientgu / VQ-Diffusion

MIT License
432 stars 43 forks source link

Questions about code in diffusion_transformer.py (q_pred) #28

Open Ed-ivan opened 1 year ago

Ed-ivan commented 1 year ago

Thanks for your excellent works and sharing your code ! while I have questions about the code: log_probs = torch.cat( [ log_add_exp(log_x_start[:,:-1,:]+log_cumprod_at, log_cumprod_bt), log_add_exp(log_x_start[:,-1:,:]+log_1_min_cumprod_ct, log_cumprod_ct) ], dim=1 ) why log_add_exp is used , looking forward your reply!

Ed-ivan commented 1 year ago

while I notice that in line 212 of diffusion_transformer.py, log_qt=self.q_pred(log_x_t,t) ,which is called by q_posterior(self,log_x_start,log_x_t,t) . I feel a little confused that why is log_x_t is used but not log_x_start

looking forward your reply!