search

LingxiaoYang2023 / DSG2024

Official pytorch repository for “Guidance with Spherical Gaussian Constraint for Conditional Diffusion”
43 stars 2 forks source link

About d_sample #6

Closed luckysyj closed 2 months ago

luckysyj commented 4 months ago

Thank you for you excellent work ! But I have a question about d_sample. In the pseudo-code, the d_sample= σ * ϵ. but in code, d_sample=x_t - x_t_mean. I don't understand what x_t_mean means. Thank you for your answer !

LingxiaoYang2023 commented 4 months ago

The term "x_tmean" refers to the posterior mean \mu\theta(x_t), where x_t represents the sample point. Therefore, x_t can be expressed x_t = x_t_mean + \sigma \epsilon. (e.g. Line 452, _guided_diffusion/gaussiandiffusion.py).