Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add masked loss implementation #589

Closed
wants to merge 7 commits into from
Closed

add masked loss implementation #589

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift click to select a range
7de0550
implement masked loss for LoRA, Textual Inversion, Dreambooth & others
Oct 9, 2023
a34222a
add back mask rescale by mean
Oct 16, 2023
b0bff65
fix mask loading when cache_latents=False or cache_latents_to_disk=True
Oct 20, 2023
610f8e0
Merge branch 'main' into masked-loss-rebase
Oct 23, 2023
45e64b1
Merge remote-tracking branch 'origin/dev' into masked-loss-rebase
Oct 23, 2023
b2adcbb
fix bugs
Oct 23, 2023
5680057
Merge branch 'dev' into masked-loss-rebase
Nov 1, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
add back mask rescale by mean
  • Loading branch information
recris committed Oct 16, 2023
commit a34222a2f033ec78d1e583392446f93664d76dd4
6 changes: 6 additions & 0 deletions library/custom_train_functions.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -458,6 458,12 @@ def apply_noise_offset(latents, noise, noise_offset, adaptive_noise_scale):


def get_latent_masks(image_masks, latent_shape, device):
# given that masks lower the average loss this will counteract the effect
factor = torch.sqrt(image_masks.mean([1, 2]))
factor = torch.where(factor != 0.0, factor, 1.0)
factor = factor.reshape(factor.shape (1,) * 2)
image_masks = image_masks / factor

masks = (
image_masks
.to(device)
Expand Down