search

XY-boy / EDiffSR

[IEEE TGRS 2024] EDiffSR: An Efficient Diffusion Probabilistic Model for Remote Sensing Image Super-Resolution
131 stars 10 forks source link

Questions about the algorithms #22

Open csyhy1986 opened 14 hours ago

csyhy1986 commented 14 hours ago

Hi, Thanks for you nice work. I think the work is based on score based model, and in my opinion, the score is the gradient of the log of the data distribution. In algorithm 1, you use f_φ to learn the score of I_t, actually I_t is the processed low resolution images which are the outputs of CPEM. While in the sampling process (i.e., Algorithm 2), the input of f_φ is v (i.e., low resolution images that are not processed), my question is, why I_t and v have same distribution?

csyhy1986 commented 13 hours ago

oops, I reviewed the codes, may be CPEM and other auxiliaries are part of f_φ? So, Algorithm 2 be may have some writing issue?