|
"""UpSample module.""" |
|
|
|
from torch import nn |
|
|
|
|
|
def up_sample(in_planes: int, out_planes: int) -> nn.Module: |
|
"""UpSample module.""" |
|
return nn.Sequential( |
|
nn.Upsample(scale_factor=2, mode="nearest"), |
|
nn.Conv2d( |
|
in_planes, out_planes * 2, kernel_size=3, stride=1, padding=1, bias=False |
|
), |
|
nn.InstanceNorm2d(out_planes * 2), |
|
nn.GLU(dim=1), |
|
) |
|
|
|
|
|
def img_up_block(in_planes: int, out_planes: int) -> nn.Module: |
|
""" |
|
Image upsample block. |
|
Mainly used to conver the 17 x 17 local feature map from Inception to 32 x 32 size. |
|
""" |
|
return nn.Sequential( |
|
nn.Upsample(scale_factor=1.9, mode="nearest"), |
|
nn.Conv2d( |
|
in_planes, out_planes * 2, kernel_size=3, stride=1, padding=1, bias=False |
|
), |
|
nn.InstanceNorm2d(out_planes * 2), |
|
nn.GLU(dim=1), |
|
) |
|
|