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

finish resnet

Browse Source
This commit is contained in:
Patrick von Platen
2022-07-01 15:40:54 +00:00
parent fa7443c899
commit 1468f754e0
1 changed files with 7 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
src/diffusers/models/unet_sde_score_estimation.py
View File

@@ -28,8 +28,7 @@ from ..modeling_utils import ModelMixin
from .attention import AttentionBlock
from .embeddings import GaussianFourierProjection, get_timestep_embedding
from .resnet import downsample_2d, upfirdn2d, upsample_2d
from .resnet import ResnetBlockBigGANppNew as ResnetBlockBigGANpp
from .resnet import ResnetBlock as ResnetNew
from .resnet import ResnetBlock
def _setup_kernel(k):
@@ -323,16 +322,6 @@ class NCSNpp(ModelMixin, ConfigMixin):
elif progressive_input == "residual":
pyramid_downsample = functools.partial(Down_sample, fir_kernel=fir_kernel, with_conv=True)
ResnetBlock = functools.partial(
ResnetBlockBigGANpp,
act=act,
dropout=dropout,
fir_kernel=fir_kernel,
init_scale=init_scale,
skip_rescale=skip_rescale,
temb_dim=nf * 4,
)
# Downsampling block
channels = num_channels
@@ -347,9 +336,8 @@ class NCSNpp(ModelMixin, ConfigMixin):
# Residual blocks for this resolution
for i_block in range(num_res_blocks):
out_ch = nf * ch_mult[i_level]
# modules.append(ResnetBlock(in_ch=in_ch, out_ch=out_ch))
modules.append(
ResnetNew(
ResnetBlock(
in_channels=in_ch,
out_channels=out_ch,
temb_channels=4 * nf,
@@ -367,9 +355,8 @@ class NCSNpp(ModelMixin, ConfigMixin):
hs_c.append(in_ch)
if i_level != self.num_resolutions - 1:
# modules.append(ResnetBlock(down=True, in_ch=in_ch))
modules.append(
ResnetNew(
ResnetBlock(
in_channels=in_ch,
temb_channels=4 * nf,
output_scale_factor=np.sqrt(2.0),
@@ -395,9 +382,8 @@ class NCSNpp(ModelMixin, ConfigMixin):
hs_c.append(in_ch)
in_ch = hs_c[-1]
# modules.append(ResnetBlock(in_ch=in_ch))
modules.append(
ResnetNew(
ResnetBlock(
in_channels=in_ch,
temb_channels=4 * nf,
output_scale_factor=np.sqrt(2.0),
@@ -408,9 +394,8 @@ class NCSNpp(ModelMixin, ConfigMixin):
)
)
modules.append(AttnBlock(channels=in_ch))
# modules.append(ResnetBlock(in_ch=in_ch))
modules.append(
ResnetNew(
ResnetBlock(
in_channels=in_ch,
temb_channels=4 * nf,
output_scale_factor=np.sqrt(2.0),
@@ -426,9 +411,8 @@ class NCSNpp(ModelMixin, ConfigMixin):
for i_level in reversed(range(self.num_resolutions)):
for i_block in range(num_res_blocks + 1):
out_ch = nf * ch_mult[i_level]
# modules.append(ResnetBlock(in_ch=in_ch + hs_c.pop(), out_ch=out_ch))
modules.append(
ResnetNew(
ResnetBlock(
in_channels=in_ch + hs_c.pop(),
out_channels=out_ch,
temb_channels=4 * nf,
@@ -470,9 +454,8 @@ class NCSNpp(ModelMixin, ConfigMixin):
raise ValueError(f"{progressive} is not a valid name")
if i_level != 0:
# modules.append(ResnetBlock(in_ch=in_ch, up=True))
modules.append(
ResnetNew(
ResnetBlock(
in_channels=in_ch,
temb_channels=4 * nf,
output_scale_factor=np.sqrt(2.0),