| """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), | |
| ) | |