search

NJU-PCALab / AddSR

71 stars 2 forks source link

Different result when testing use the same image #16

Open rainsoulsrx opened 5 days ago

rainsoulsrx commented 5 days ago

Hi, I test the image you display here https://imgsli.com/MjUyNTgx, but the result is different from yours. why?

CSRuiXie commented 4 days ago

Thank you for your attention to our work. This may be caused by the random seed leading to different initial noisy states (i.e., at t=999). You can change the seed to generate different results.

rainsoulsrx commented 4 days ago

Thank you for your attention to our work. This may be caused by the random seed leading to different initial noisy states (i.e., at t=999). You can change the seed to generate different results.

which seed do you use in the displayed image?

CSRuiXie commented 4 days ago

We used the default setting, which set the seed to None.

rainsoulsrx commented 4 days ago

We used the default setting, which set the seed to None.

I use default setting, too. I check the seed is None. But the result are very different

CSRuiXie commented 3 days ago

Due to the stochasticity of DM-based models, the results can look very different. If the results you get are not satisfactory, you can change the seed to try other results. This phenomenon is discussed in CCSR, where you can find more detailed discussions. With this in mind, this issue is important for using DM-based models in SR tasks, and we will consider it in our future plans.