Skip to content

Fix _upsample_2d #535

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

Merged
ydshieh merged 1 commit into from
Sep 19, 2022
Merged

Fix _upsample_2d #535

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
3 changes: 1 addition & 2 deletions src/diffusers/models/resnet.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -149,7 +149,6 @@ def _upsample_2d(self, x, weight=None, kernel=None, factor=2, gain=1):

stride = (factor, factor)
# Determine data dimensions.
stride = [1, 1, factor, factor]
Copy link
Contributor Author

@ydshieh ydshieh Sep 16, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

F.conv_transpose2d should have a single int or a tuple of 2 integers for stride, not 4 integers.

output_shape = ((x.shape[2] - 1) * factor + convH, (x.shape[3] - 1) * factor + convW)
output_padding = (
output_shape[0] - (x.shape[2] - 1) * stride[0] - convH,
Expand All @@ -161,7 +160,7 @@ def _upsample_2d(self, x, weight=None, kernel=None, factor=2, gain=1):

# Transpose weights.
weight = torch.reshape(weight, (num_groups, -1, inC, convH, convW))
weight = weight[..., ::-1, ::-1].permute(0, 2, 1, 3, 4)
weight = torch.flip(weight, dims=[3, 4]).permute(0, 2, 1, 3, 4)
Copy link
Contributor Author

@ydshieh ydshieh Sep 16, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Without the fix, we got

ValueError: step must be greater than zero

The syntax ::-1 is for TF tensors, not for torch tensors.

weight = torch.reshape(weight, (num_groups * inC, -1, convH, convW))

x = F.conv_transpose2d(x, weight, stride=stride, output_padding=output_padding, padding=0)
Expand Down