Open jueqiw opened 1 year ago
I have the sam question. When I campare equation of deterministic DDIM sampling in the Null-text Inversion paper, it is different with DDIM & DiffuisonCLIP & Diffusion models beat GAN papers.
same question here, thought it would be a typo
It is a typo!
Hi,
Thanks for this wonderful work! I have a question about the equation of deterministic DDIM sampling in the Null-text Inversion paper. Based on my understanding, deterministic DDIM sampling is to set the $\sigmat=0$ for equation 12 in the DDIM paper. It should be the following equation: $z{t-1} = \sqrt{\alpha_{t-1}}\left(\frac{z_t - \sqrt{1-\alphat}\cdot\epsilon\theta(z_t,\mathcal{C},t)}{\sqrt{\alphat}}\right) + \sqrt{1-\alpha{t-1}}\cdot\epsilon_\theta(zt, \mathcal{C}, t)$ If you rewrite this equation into the Null-text Inversion paper version, it should be: $z{t-1} = \sqrt{\frac{\alpha_{t-1}}{\alpha_t}}zt + \sqrt{\alpha{t-1}}\left(\sqrt{\frac{1}{\alpha_{t-1}} - 1}- \sqrt{\frac{1}{\alphat}-1} \right)\cdot\epsilon\theta(z_t, \mathcal{C}, t)$ Which is different from it the Null-text Inversion paper.
It may be my understanding is wrong. I would very much appreciate it if you could point me in the right direction!
Thanks,
Jueqi