Implementing cycleGAN for unsupervised image to image domain translation. See https://junyanz.github.io/CycleGAN/.

Building Blocks

class AutoTransConv[source]

AutoTransConv(n_in, n_out, ks, stride, bias=True) :: Module

Automatic padding of transpose convolution for input-output feature size

trans_conv_norm_relu[source]

trans_conv_norm_relu(n_in, n_out, norm_layer, bias, ks=3, stride=2)

Transpose convolutional layer

pad_conv_norm_relu[source]

pad_conv_norm_relu(n_in, n_out, norm_layer, padding_mode='zeros', pad=1, ks=3, stride=1, activ=True, bias=True)

Adding ability to specify different paddings to convolutional layer

conv_norm_relu[source]

conv_norm_relu(n_in, n_out, norm_layer=None, ks=3, bias:bool=True, pad=1, stride=1, activ=True, a=0.2)

Convolutional layer

class ResBlock[source]

ResBlock(dim, padding_mode, bias, dropout, norm_layer='InstanceNorm2d') :: Module

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes::

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call :meth:to, etc.

Generator

generator[source]

generator(n_in, n_out, n_f=64, norm_layer=None, dropout=0.0, n_blocks=6, pad_mode='reflection')

Generator that maps an input of one domain to the other

generator(3,10)

Sequential(
  (0): ReflectionPad2d((0, 0, 0, 0))
  (1): AutoConv(
    (conv): Conv2d(3, 64, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3), padding_mode=reflection)
  )
  (2): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (3): ReLU(inplace=True)
  (4): AutoConv(
    (conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
  )
  (5): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (6): ReLU(inplace=True)
  (7): AutoConv(
    (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
  )
  (8): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (9): ReLU(inplace=True)
  (10): ResBlock(
    (xb): Sequential(
      (0): ReflectionPad2d((0, 0, 0, 0))
      (1): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      (3): ReLU(inplace=True)
      (4): ReflectionPad2d((0, 0, 0, 0))
      (5): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    )
  )
  (11): ResBlock(
    (xb): Sequential(
      (0): ReflectionPad2d((0, 0, 0, 0))
      (1): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      (3): ReLU(inplace=True)
      (4): ReflectionPad2d((0, 0, 0, 0))
      (5): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    )
  )
  (12): ResBlock(
    (xb): Sequential(
      (0): ReflectionPad2d((0, 0, 0, 0))
      (1): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      (3): ReLU(inplace=True)
      (4): ReflectionPad2d((0, 0, 0, 0))
      (5): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    )
  )
  (13): ResBlock(
    (xb): Sequential(
      (0): ReflectionPad2d((0, 0, 0, 0))
      (1): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      (3): ReLU(inplace=True)
      (4): ReflectionPad2d((0, 0, 0, 0))
      (5): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    )
  )
  (14): ResBlock(
    (xb): Sequential(
      (0): ReflectionPad2d((0, 0, 0, 0))
      (1): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      (3): ReLU(inplace=True)
      (4): ReflectionPad2d((0, 0, 0, 0))
      (5): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    )
  )
  (15): ResBlock(
    (xb): Sequential(
      (0): ReflectionPad2d((0, 0, 0, 0))
      (1): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      (3): ReLU(inplace=True)
      (4): ReflectionPad2d((0, 0, 0, 0))
      (5): AutoConv(
        (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
      )
      (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    )
  )
  (16): AutoTransConv(
    (conv): ConvTranspose2d(256, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
  )
  (17): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (18): ReLU()
  (19): AutoTransConv(
    (conv): ConvTranspose2d(128, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
  )
  (20): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (21): ReLU()
  (22): ReflectionPad2d((3, 3, 3, 3))
  (23): Conv2d(64, 10, kernel_size=(7, 7), stride=(1, 1))
  (24): Tanh()
)

Discriminator

discriminator[source]

discriminator(c_in, n_f, n_layers, norm_layer=None, sigmoid=False)

Discrminator to classify input as belonging to one class or the other

discriminator(3, 6, 3)

Sequential(
  (0): Conv2d(3, 6, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
  (1): LeakyReLU(negative_slope=0.2, inplace=True)
  (2): Conv2d(6, 12, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
  (3): InstanceNorm2d(12, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (4): LeakyReLU(negative_slope=0.2, inplace=True)
  (5): Conv2d(12, 24, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
  (6): InstanceNorm2d(24, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (7): LeakyReLU(negative_slope=0.2, inplace=True)
  (8): Conv2d(24, 48, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
  (9): InstanceNorm2d(48, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
  (10): LeakyReLU(negative_slope=0.2, inplace=True)
  (11): Conv2d(48, 1, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
)

Loss and Trainer

class cycleGAN[source]

cycleGAN(c_in, c_out, n_f=64, disc_layers=3, gen_blocks=6, drop=0.0, norm_layer=None, sigmoid=False) :: Module

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes::

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call :meth:to, etc.

class DynamicLoss[source]

DynamicLoss(loss_fn) :: Module

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes::

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call :meth:to, etc.

class CycleGANLoss[source]

CycleGANLoss(model, loss_fn='mse_loss', la=10.0, lb=10, lid=0.5) :: Module

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes::

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call :meth:to, etc.

CycleGAN loss is composed of 3 parts:

  1. Identity, an image that has gone through its own domain should remain the same
  2. Generator, the output images should fool the discriminator into thinking they belong to that class
  3. Cyclical loss, and image that has been mapped to the other domain then mapped back to itself should resemble the original input

class cycleGANTrainer[source]

cycleGANTrainer() :: Callback

Trainer to sequence timing of training both the discriminator as well as the generator's

Training

cgan = cycleGAN(3, 3, gen_blocks=9)
learn = get_learner(cgan, loss=CycleGANLoss(cgan))
run = get_runner(learn, [cycleGANTrainer()])

cgan

cycleGAN(
  (a_discriminator): Sequential(
    (0): Conv2d(3, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (1): LeakyReLU(negative_slope=0.2, inplace=True)
    (2): Conv2d(64, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (3): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (4): LeakyReLU(negative_slope=0.2, inplace=True)
    (5): Conv2d(128, 256, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (7): LeakyReLU(negative_slope=0.2, inplace=True)
    (8): Conv2d(256, 512, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
    (9): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (10): LeakyReLU(negative_slope=0.2, inplace=True)
    (11): Conv2d(512, 1, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
  )
  (b_discriminator): Sequential(
    (0): Conv2d(3, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (1): LeakyReLU(negative_slope=0.2, inplace=True)
    (2): Conv2d(64, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (3): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (4): LeakyReLU(negative_slope=0.2, inplace=True)
    (5): Conv2d(128, 256, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (7): LeakyReLU(negative_slope=0.2, inplace=True)
    (8): Conv2d(256, 512, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
    (9): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (10): LeakyReLU(negative_slope=0.2, inplace=True)
    (11): Conv2d(512, 1, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
  )
  (generate_a): Sequential(
    (0): ReflectionPad2d((0, 0, 0, 0))
    (1): AutoConv(
      (conv): Conv2d(3, 64, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3), padding_mode=reflection)
    )
    (2): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (3): ReLU(inplace=True)
    (4): AutoConv(
      (conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
    )
    (5): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (6): ReLU(inplace=True)
    (7): AutoConv(
      (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
    )
    (8): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (9): ReLU(inplace=True)
    (10): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (11): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (12): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (13): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (14): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (15): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (16): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (17): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (18): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (19): AutoTransConv(
      (conv): ConvTranspose2d(256, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
    )
    (20): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (21): ReLU()
    (22): AutoTransConv(
      (conv): ConvTranspose2d(128, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
    )
    (23): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (24): ReLU()
    (25): ReflectionPad2d((3, 3, 3, 3))
    (26): Conv2d(64, 3, kernel_size=(7, 7), stride=(1, 1))
    (27): Tanh()
  )
  (generate_b): Sequential(
    (0): ReflectionPad2d((0, 0, 0, 0))
    (1): AutoConv(
      (conv): Conv2d(3, 64, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3), padding_mode=reflection)
    )
    (2): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (3): ReLU(inplace=True)
    (4): AutoConv(
      (conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
    )
    (5): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (6): ReLU(inplace=True)
    (7): AutoConv(
      (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
    )
    (8): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (9): ReLU(inplace=True)
    (10): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (11): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (12): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (13): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (14): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (15): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (16): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (17): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (18): ResBlock(
      (xb): Sequential(
        (0): ReflectionPad2d((0, 0, 0, 0))
        (1): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
        (3): ReLU(inplace=True)
        (4): ReflectionPad2d((0, 0, 0, 0))
        (5): AutoConv(
          (conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), padding_mode=reflection)
        )
        (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
      )
    )
    (19): AutoTransConv(
      (conv): ConvTranspose2d(256, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
    )
    (20): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (21): ReLU()
    (22): AutoTransConv(
      (conv): ConvTranspose2d(128, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
    )
    (23): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (24): ReLU()
    (25): ReflectionPad2d((3, 3, 3, 3))
    (26): Conv2d(64, 3, kernel_size=(7, 7), stride=(1, 1))
    (27): Tanh()
  )
)