def __init__(self, input_channel, output_channel, ID):
super(UNet_up_block2, self).__init__()
kernel_size = 3
self.ID = ID
self.up_sampling = torch.nn.ConvTranspose2d(input_channel,
input_channel,
4,
stride=(2, 2),
padding=(1, 1))
self.conv1 = torch.nn.Conv2d(input_channel,
output_channel,
kernel_size,
stride=(1, 1),
padding=(1, 1),
bias=False)
self.bn1 = Norm(output_channel)
self.conv2 = torch.nn.Conv2d(output_channel,
output_channel,
kernel_size,
stride=(1, 1),
padding=(1, 1),
bias=False)
self.bn2 = Norm(output_channel)
self.conv3 = torch.nn.Conv2d(output_channel,
output_channel,
kernel_size,
stride=(1, 1),
padding=(1, 1),
bias=False)
self.bn3 = Norm(output_channel)
def __init__(self, input_channel, output_channel, down_sample):
super(UNet_down_block, self).__init__()
kernel_size = 3
self.conv1 = torch.nn.Conv2d(input_channel,
output_channel,
kernel_size,
stride=(1, 1),
padding=(1, 1),
bias=False)
self.bn1 = Norm(output_channel)
self.conv2 = torch.nn.Conv2d(output_channel,
output_channel,
kernel_size,
stride=(1, 1),
padding=(1, 1),
bias=False)
self.bn2 = Norm(output_channel)
self.conv3 = torch.nn.Conv2d(output_channel,
output_channel,
kernel_size,
stride=(1, 1),
padding=(1, 1),
bias=False)
self.bn3 = Norm(output_channel)
self.down_sampling = torch.nn.Conv2d(input_channel,
input_channel,
kernel_size,
stride=(2, 2),
padding=(1, 1),
bias=False)
self.down_sample = down_sample
def __init__(self):
super(UNet, self).__init__()
# self.opts = opts
input_channel_number =160
output_channel_number =160
# kernel_size = opts.kernel_size # we could change this later
kernel_size = 3
# self.linear1 = torch.nn.Sequential(*self.lin_tan_drop(400 * 520, 400 * 520, 64))
# self.linear2 = torch.nn.Sequential(*self.lin_tan_drop(200 * 260, 200 * 260, 64))
#
# self.linear1 = torch.nn.Sequential(*self.lin_tan_drop(220 * 220, 229 * 220, 64))
# self.linear2 = torch.nn.Sequential(*self.lin_tan_drop(110 * 110, 110 * 110, 64))
# self.linear1 = torch.nn.Sequential(*self.lin_tan_drop(400 * 520, 400 * 520, 64))
# self.linear2 = torch.nn.Sequential(*self.lin_tan_drop(200 * 260, 200 * 260, 64))
# self.linear3 = torch.nn.Sequential(*self.lin_tan_drop(100 * 130, 100 * 130, 64))
# Encoder network
self.down_block1 = UNet_down_block(input_channel_number, 64, False) # 64*520
self.down_block2 = UNet_down_block(64, 128, True) # 64*520
self.down_block3 = UNet_down_block(128, 256, True) # 64*260
# bottom convolution
self.mid_conv1 = torch.nn.Conv2d(256, 256, kernel_size, padding=(1, 1), bias=False)# 64*260
self.bn1 = Norm(256)
self.mid_conv2 = torch.nn.Conv2d(256, 256, kernel_size, padding=(1, 1), bias=False)# 64*260
self.bn2 = Norm(256)
self.mid_conv3 = torch.nn.Conv2d(256, 256, kernel_size, padding=(1, 1), bias=False) #, dilation=4 # 64*260
self.bn3 = Norm(256)
self.mid_conv4 = torch.nn.Conv2d(256, 256, kernel_size, padding=(1, 1), bias=False)# 64*260
self.bn4 = Norm(256)
self.mid_conv5 = torch.nn.Conv2d(256, 256, kernel_size, padding=(1, 1), bias=False)# 64*260
self.bn5 = Norm(256)
# Decoder network
# self.up_block2 = UNet_up_block(128, 256, 128, 1)# 64*520
# self.up_block3 = UNet_up_block(64, 128, 64, 1)# 64*520
#
self.up_block21 = UNet_up_block2(128, 128, 256, 128, 1)# 64*520
self.up_block31 = UNet_up_block2(64, 64, 128, 64, 1)# 64*520
# Final output
self.last_conv1 = torch.nn.Conv2d(64, 64, 3, padding=(1, 1), bias=False)# 64*520
self.last_bn = Norm(64) #
self.last_conv2 = torch.nn.Conv2d(64, output_channel_number, 3, padding=(1, 1))# 64*520
self.last_bn2 = Norm(output_channel_number) # 64*520
self.softplus = torch.nn.Softplus(beta=5, threshold=100)
self.relu = torch.nn.ReLU()
self.tanhshrink = torch.nn.Tanhshrink()
self.tanh = torch.nn.Tanh()
def __init__(self):
super(UNet, self).__init__()
# self.opts = opts
input_channel_number = 1
output_channel_number = 1
kernel_size = 3
# Encoder network
self.down_block1 = UNet_down_block(input_channel_number, 64,
False) # 64*520
self.down_block2 = UNet_down_block(64, 128, True) # 64*520
self.down_block3 = UNet_down_block(128, 256, True) # 64*260
self.down_block4 = UNet_down_block(256, 512, True) # 64*260
# bottom convolution
self.mid_conv1 = torch.nn.Conv2d(512,
512,
kernel_size,
padding=(1, 1),
bias=False) # 64*260
self.bn1 = Norm(512)
self.mid_conv2 = torch.nn.Conv2d(512,
512,
kernel_size,
padding=(1, 1),
bias=False) # 64*260
self.bn2 = Norm(512)
self.mid_conv3 = torch.nn.Conv2d(512,
512,
kernel_size,
padding=(1, 1),
bias=False) #, dilation=4 # 64*260
self.bn3 = Norm(512)
self.mid_conv4 = torch.nn.Conv2d(512,
512,
kernel_size,
padding=(1, 1),
bias=False) # 64*260
self.bn4 = Norm(512)
self.mid_conv5 = torch.nn.Conv2d(512,
512,
kernel_size,
padding=(1, 1),
bias=False) # 64*260
self.bn5 = Norm(512)
# Decoder network
self.up_block2 = UNet_up_block2(512, 256, 1) # 64*520
self.up_block3 = UNet_up_block2(256, 64, 1) # 64*520
# self.up_block4 = UNet_up_block2(128, 64, 1)# 64*520
# Final output
self.last_conv1 = torch.nn.Conv2d(64,
64,
3,
padding=(1, 1),
bias=False) # 64*520
self.last_bn = Norm(1) #
self.last_conv2 = torch.nn.Conv2d(64,
output_channel_number,
3,
padding=(1, 1)) # 64*520
self.last_bn2 = Norm(output_channel_number) # 64*520
self.softplus = torch.nn.Softplus(beta=5, threshold=100)
self.relu = torch.nn.ReLU()
self.tanhshrink = torch.nn.Tanhshrink()
self.tanh = torch.nn.Tanh()