Open Shi5013 opened 2 weeks ago
Please refer to: https://arxiv.org/pdf/2208.09392. https://github.com/arpitbansal297/Cold-Diffusion-Models Cold Diffusion: Inverting Arbitrary Image Transforms Without Noise. In my opinion, the time step should be adjusted according to the different datasets.
Hi!
I notice that you have a parameter
time_steps = 50 # Time step - number of noise stacks
indriver.py
. In traditional DDPM model, the time step uausally be setted hundred or thousand,like 1000. I want to know their difference?If I use a different dataset,shoud I change it?