Open ZengJia-Notes opened 1 year ago
We do not provide the version with the 'predict_xstart=True'.
To enable its functionality, you might modify ./utils/diffusion_utils.py/denoising_step
Thanks
I'm almost clueless about where to start. How can I go about modifying the code? Thank U
I have trained an i-ddpm model with predict_xstart = True, which instructs the diffusion model to output the value of x0.
Does DiffStyle also work when the model is outputting x0?
If not, which parts of the code should I modify to enable its functionality?
Thank you.