Hello. In your code, I think that there are four types of timesteps_stu: 249, 499, 749, 999. However, in timesteps_stu, there are only three choices of timesteps_stu: 499, 749 and 999, and in line 1349 in train_addsr.py, indices_1 = torch.nonzero(timesteps_stu != 249).squeeze(), does not have any practical functions. And what is the function of the timesteps_stu_ in the "second steps" in your code?
Besides, I also want to know how did you implement the PSR module in your paper. In the PSR module, there seems to be four steps, where each step contains the controlnet and SD part. You only have two steps. I only see the variable p. If p>0.5, then the steps may be 999 in the first step and there would not the second step?
Hello. In your code, I think that there are four types of timesteps_stu: 249, 499, 749, 999. However, in timesteps_stu, there are only three choices of timesteps_stu: 499, 749 and 999, and in line 1349 in train_addsr.py,
indices_1 = torch.nonzero(timesteps_stu != 249).squeeze(), does not have any practical functions. And what is the function of thetimesteps_stu_in the "second steps" in your code?Besides, I also want to know how did you implement the PSR module in your paper. In the PSR module, there seems to be four steps, where each step contains the controlnet and SD part. You only have two steps. I only see the variable
p. Ifp>0.5, then the steps may be 999 in the first step and there would not the second step?