Open Maydaybye opened 5 months ago
Generating new noise here to assign to sample_img, won't it be inconsistent with the learned distribution, thereby affecting the result?
The noise you mentioned is added according to the forward formula of diffusion models (the same as the training process), which is still consistent with the learned distribution.
Generating new noise here to assign to sample_img, won't it be inconsistent with the learned distribution, thereby affecting the result?