def __init__(self, in_channels, out_channels, expansion=6, stride=2):
super().__init__()
self.use_shortcut = stride == 1 and in_channels == out_channels
expand_channels = in_channels * expansion
self.block = nn.Sequential(
# pw
ConvBNReLU(
in_channels=in_channels,
out_channels=expand_channels,
kernel_size=1,
bias_attr=False),
# dw
ConvBNReLU(
in_channels=expand_channels,
out_channels=expand_channels,
kernel_size=3,
stride=stride,
padding=1,
groups=expand_channels,
bias_attr=False),
# pw-linear
ConvBN(
in_channels=expand_channels,
out_channels=out_channels,
kernel_size=1,
bias_attr=False))
def __init__(self, in_channels):
super().__init__()
self.double_conv = nn.Sequential(ConvBNReLU(in_channels, 64, 3),
ConvBNReLU(64, 64, 3))
down_channels = [[64, 128], [128, 256], [256, 512], [512, 512]]
self.down_sample_list = nn.LayerList([
self.down_sampling(channel[0], channel[1])
for channel in down_channels
])
def __init__(self, in_channels, num_image, first_concat=True):
super().__init__()
bate = 1 if first_concat == False else num_image
self.double_conv = nn.Sequential(
ConvBNReLU(bate*in_channels, 64, 3), ConvBNReLU(64, 64, 3))
down_channels = [[64, 128], [128, 256], [256, 512], [512, 512]]
self.down_sample_list = nn.LayerList([
self.down_sampling(channel[0], channel[1])
for channel in down_channels
])
def __init__(self,
in_channels,
out_channels,
align_corners,
use_deconv=False):
super().__init__()
self.align_corners = align_corners
self.use_deconv = use_deconv
if self.use_deconv:
self.deconv = nn.Conv2DTranspose(in_channels,
out_channels // 2,
kernel_size=2,
stride=2,
padding=0)
in_channels = in_channels + out_channels // 2
else:
in_channels *= 2
self.double_conv = nn.Sequential(
ConvBNReLU(in_channels, out_channels, 3),
ConvBNReLU(out_channels, out_channels, 3))
def __init__(self, high_in_channels, low_in_channels, out_channels,
align_corners):
super().__init__()
# Only depth-wise conv
self.dwconv = ConvBNReLU(
in_channels=low_in_channels,
out_channels=out_channels,
kernel_size=3,
padding=1,
groups=128,
bias_attr=False)
self.conv_low_res = ConvBN(out_channels, out_channels, 1)
self.conv_high_res = ConvBN(high_in_channels, out_channels, 1)
self.align_corners = align_corners
def __init__(self, in_channels=3, dw_channels1=32, dw_channels2=48, out_channels=64):
super(LearningToDownsample, self).__init__()
self.conv_bn_relu = ConvBNReLU(
in_channels=2*in_channels, out_channels=dw_channels1, kernel_size=3, stride=2)
self.dsconv_bn_relu1 = SeparableConvBNReLU(
in_channels=dw_channels1,
out_channels=dw_channels2,
kernel_size=3,
stride=2,
padding=1)
self.dsconv_bn_relu2 = SeparableConvBNReLU(
in_channels=dw_channels2,
out_channels=out_channels,
kernel_size=3,
stride=2,
padding=1)
def down_sampling(self, in_channels, out_channels):
modules = []
modules.append(nn.MaxPool2D(kernel_size=2, stride=2))
modules.append(ConvBNReLU(in_channels, out_channels, 3))
modules.append(ConvBNReLU(out_channels, out_channels, 3))
return nn.Sequential(*modules)