AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

Revert RePaint scheduler 'fix' (#10644)

Browse Source 
Co-authored-by: Giuseppe Catalano <giuseppelorenzo.catalano@unito.it>
This commit is contained in:
Giuseppe Catalano
2025-01-27 22:16:45 +01:00
committed by GitHub
parent e89ab5bc26
commit fb42066489
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/diffusers/schedulers/scheduling_repaint.py
View File

@@ -319,7 +319,11 @@ class RePaintScheduler(SchedulerMixin, ConfigMixin):
prev_unknown_part = alpha_prod_t_prev**0.5 * pred_original_sample + pred_sample_direction + variance
# 8. Algorithm 1 Line 5 https://arxiv.org/pdf/2201.09865.pdf
prev_known_part = (alpha_prod_t_prev**0.5) * original_image + (1 - alpha_prod_t_prev) * noise
# The computation reported in Algorithm 1 Line 5 is incorrect. Line 5 refers to formula (8a) of the same paper,
# which tells to sample from a Gaussian distribution with mean "(alpha_prod_t_prev**0.5) * original_image"
# and variance "(1 - alpha_prod_t_prev)". This means that the standard Gaussian distribution "noise" should be
# scaled by the square root of the variance (as it is done here), however Algorithm 1 Line 5 tells to scale by the variance.
prev_known_part = (alpha_prod_t_prev**0.5) * original_image + ((1 - alpha_prod_t_prev) ** 0.5) * noise
# 9. Algorithm 1 Line 8 https://arxiv.org/pdf/2201.09865.pdf
pred_prev_sample = mask * prev_known_part + (1.0 - mask) * prev_unknown_part