Skip to content

Remove the usage of numpy in up/down sample_2d #503

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
merged 4 commits into from
Sep 16, 2022
Merged

Remove the usage of numpy in up/down sample_2d #503

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f38dbda
Fix PT up/down sample_2d
ydshieh Sep 13, 2022
4c05bed
empty commit
ydshieh Sep 13, 2022
7a1ebbf
style
ydshieh Sep 13, 2022
720f034
style
ydshieh Sep 13, 2022
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
31 changes: 14 additions & 17 deletions src/diffusers/models/resnet.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,5 @@
from functools import partial

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
Expand Down Expand Up @@ -134,10 +133,10 @@ def _upsample_2d(self, x, weight=None, kernel=None, factor=2, gain=1):
kernel = [1] * factor

# setup kernel
kernel = np.asarray(kernel, dtype=np.float32)
kernel = torch.tensor(kernel, dtype=torch.float32)
if kernel.ndim == 1:
kernel = np.outer(kernel, kernel)
kernel /= np.sum(kernel)
kernel = torch.outer(kernel, kernel)
kernel /= torch.sum(kernel)

kernel = kernel * (gain * (factor**2))

Expand Down Expand Up @@ -219,10 +218,10 @@ def _downsample_2d(self, x, weight=None, kernel=None, factor=2, gain=1):
kernel = [1] * factor

# setup kernel
kernel = np.asarray(kernel, dtype=np.float32)
kernel = torch.tensor(kernel, dtype=torch.float32)
if kernel.ndim == 1:
kernel = np.outer(kernel, kernel)
kernel /= np.sum(kernel)
kernel = torch.outer(kernel, kernel)
kernel /= torch.sum(kernel)

kernel = kernel * gain

Expand Down Expand Up @@ -391,16 +390,14 @@ def upsample_2d(x, kernel=None, factor=2, gain=1):
if kernel is None:
kernel = [1] * factor

kernel = np.asarray(kernel, dtype=np.float32)
kernel = torch.tensor(kernel, dtype=torch.float32)
if kernel.ndim == 1:
kernel = np.outer(kernel, kernel)
kernel /= np.sum(kernel)
kernel = torch.outer(kernel, kernel)
kernel /= torch.sum(kernel)

kernel = kernel * (gain * (factor**2))
p = kernel.shape[0] - factor
return upfirdn2d_native(
x, torch.tensor(kernel, device=x.device), up=factor, pad=((p + 1) // 2 + factor - 1, p // 2)
)
return upfirdn2d_native(x, kernel.to(device=x.device), up=factor, pad=((p + 1) // 2 + factor - 1, p // 2))


def downsample_2d(x, kernel=None, factor=2, gain=1):
Expand All @@ -425,14 +422,14 @@ def downsample_2d(x, kernel=None, factor=2, gain=1):
if kernel is None:
kernel = [1] * factor

kernel = np.asarray(kernel, dtype=np.float32)
kernel = torch.tensor(kernel, dtype=torch.float32)
if kernel.ndim == 1:
kernel = np.outer(kernel, kernel)
kernel /= np.sum(kernel)
kernel = torch.outer(kernel, kernel)
kernel /= torch.sum(kernel)

kernel = kernel * gain
p = kernel.shape[0] - factor
return upfirdn2d_native(x, torch.tensor(kernel, device=x.device), down=factor, pad=((p + 1) // 2, p // 2))
return upfirdn2d_native(x, kernel.to(device=x.device), down=factor, pad=((p + 1) // 2, p // 2))


def upfirdn2d_native(input, kernel, up=1, down=1, pad=(0, 0)):
Expand Down