169 lines
6.7 KiB
Python
169 lines
6.7 KiB
Python
import torch
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
import numpy as np
|
|
import sys
|
|
|
|
import defenses.smoothing as smoothing
|
|
|
|
class ResidualBlock(nn.Module):
|
|
"""Residual Block with instance normalization."""
|
|
def __init__(self, dim_in, dim_out):
|
|
super(ResidualBlock, self).__init__()
|
|
self.main = nn.Sequential(
|
|
nn.Conv2d(dim_in, dim_out, kernel_size=3, stride=1, padding=1, bias=False),
|
|
nn.InstanceNorm2d(dim_out, affine=True, track_running_stats=True),
|
|
nn.ReLU(inplace=True),
|
|
nn.Conv2d(dim_out, dim_out, kernel_size=3, stride=1, padding=1, bias=False),
|
|
nn.InstanceNorm2d(dim_out, affine=True, track_running_stats=True))
|
|
|
|
def forward(self, x):
|
|
return x + self.main(x)
|
|
|
|
|
|
class Generator(nn.Module):
|
|
"""Generator network."""
|
|
def __init__(self, conv_dim=64, c_dim=5, repeat_num=6):
|
|
super(Generator, self).__init__()
|
|
|
|
layers = []
|
|
layers.append(nn.Conv2d(3+c_dim, conv_dim, kernel_size=7, stride=1, padding=3, bias=False))
|
|
layers.append(nn.InstanceNorm2d(conv_dim, affine=True, track_running_stats=True))
|
|
layers.append(nn.ReLU(inplace=True))
|
|
|
|
# Down-sampling layers.
|
|
curr_dim = conv_dim
|
|
for i in range(2):
|
|
layers.append(nn.Conv2d(curr_dim, curr_dim*2, kernel_size=4, stride=2, padding=1, bias=False))
|
|
layers.append(nn.InstanceNorm2d(curr_dim*2, affine=True, track_running_stats=True))
|
|
layers.append(nn.ReLU(inplace=True))
|
|
curr_dim = curr_dim * 2
|
|
|
|
# Bottleneck layers.
|
|
for i in range(repeat_num):
|
|
layers.append(ResidualBlock(dim_in=curr_dim, dim_out=curr_dim))
|
|
|
|
# Up-sampling layers.
|
|
for i in range(2):
|
|
layers.append(nn.ConvTranspose2d(curr_dim, curr_dim//2, kernel_size=4, stride=2, padding=1, bias=False))
|
|
layers.append(nn.InstanceNorm2d(curr_dim//2, affine=True, track_running_stats=True))
|
|
layers.append(nn.ReLU(inplace=True))
|
|
curr_dim = curr_dim // 2
|
|
|
|
layers.append(nn.Conv2d(curr_dim, 3, kernel_size=7, stride=1, padding=3, bias=False))
|
|
layers.append(nn.Tanh())
|
|
self.main = nn.Sequential(*layers)
|
|
|
|
def forward(self, x, c):
|
|
# Replicate spatially and concatenate domain information.
|
|
# Note that this type of label conditioning does not work at all if we use reflection padding in Conv2d.
|
|
# This is because instance normalization ignores the shifting (or bias) effect.
|
|
c = c.view(c.size(0), c.size(1), 1, 1)
|
|
c = c.repeat(1, 1, x.size(2), x.size(3))
|
|
x = torch.cat([x, c], dim=1)
|
|
|
|
feature_maps = []
|
|
|
|
# Get intermediate feature maps
|
|
for layer in self.main:
|
|
# print('test')
|
|
x = layer(x)
|
|
feature_maps.append(x)
|
|
# print(len(feature_maps))
|
|
|
|
return x, feature_maps
|
|
|
|
def forward_blur(self, x, c, blur_layer):
|
|
c = c.view(c.size(0), c.size(1), 1, 1)
|
|
c = c.repeat(1, 1, x.size(2), x.size(3))
|
|
x = blur_layer(x)
|
|
x = torch.cat([x, c], dim=1)
|
|
|
|
feature_maps = []
|
|
# Get intermediate feature maps
|
|
for layer in self.main:
|
|
x = layer(x)
|
|
feature_maps.append(x)
|
|
|
|
return x, feature_maps
|
|
|
|
class Discriminator(nn.Module):
|
|
"""Discriminator network with PatchGAN."""
|
|
def __init__(self, image_size=128, conv_dim=64, c_dim=5, repeat_num=6):
|
|
super(Discriminator, self).__init__()
|
|
layers = []
|
|
layers.append(nn.Conv2d(3, conv_dim, kernel_size=4, stride=2, padding=1))
|
|
layers.append(nn.LeakyReLU(0.01))
|
|
|
|
curr_dim = conv_dim
|
|
for i in range(1, repeat_num):
|
|
layers.append(nn.Conv2d(curr_dim, curr_dim*2, kernel_size=4, stride=2, padding=1))
|
|
layers.append(nn.LeakyReLU(0.01))
|
|
curr_dim = curr_dim * 2
|
|
|
|
kernel_size = int(image_size / np.power(2, repeat_num))
|
|
self.main = nn.Sequential(*layers)
|
|
self.conv1 = nn.Conv2d(curr_dim, 1, kernel_size=3, stride=1, padding=1, bias=False)
|
|
self.conv2 = nn.Conv2d(curr_dim, c_dim, kernel_size=kernel_size, bias=False)
|
|
|
|
def forward(self, x):
|
|
h = self.main(x)
|
|
out_src = self.conv1(h)
|
|
out_cls = self.conv2(h)
|
|
return out_src, out_cls.view(out_cls.size(0), out_cls.size(1))
|
|
|
|
class AvgBlurGenerator(nn.Module):
|
|
"""Generator network."""
|
|
def __init__(self, conv_dim=64, c_dim=5, repeat_num=6):
|
|
super(AvgBlurGenerator, self).__init__()
|
|
|
|
layers = []
|
|
layers.append(nn.Conv2d(3+c_dim, conv_dim, kernel_size=7, stride=1, padding=3, bias=False))
|
|
layers.append(nn.InstanceNorm2d(conv_dim, affine=True, track_running_stats=True))
|
|
layers.append(nn.ReLU(inplace=True))
|
|
|
|
# Down-sampling layers.
|
|
curr_dim = conv_dim
|
|
for i in range(2):
|
|
layers.append(nn.Conv2d(curr_dim, curr_dim*2, kernel_size=4, stride=2, padding=1, bias=False))
|
|
layers.append(nn.InstanceNorm2d(curr_dim*2, affine=True, track_running_stats=True))
|
|
layers.append(nn.ReLU(inplace=True))
|
|
curr_dim = curr_dim * 2
|
|
|
|
# Bottleneck layers.
|
|
for i in range(repeat_num):
|
|
layers.append(ResidualBlock(dim_in=curr_dim, dim_out=curr_dim))
|
|
|
|
# Up-sampling layers.
|
|
for i in range(2):
|
|
layers.append(nn.ConvTranspose2d(curr_dim, curr_dim//2, kernel_size=4, stride=2, padding=1, bias=False))
|
|
layers.append(nn.InstanceNorm2d(curr_dim//2, affine=True, track_running_stats=True))
|
|
layers.append(nn.ReLU(inplace=True))
|
|
curr_dim = curr_dim // 2
|
|
|
|
layers.append(nn.Conv2d(curr_dim, 3, kernel_size=7, stride=1, padding=3, bias=False))
|
|
layers.append(nn.Tanh())
|
|
self.main = nn.Sequential(*layers)
|
|
|
|
layers_preproc = []
|
|
# layers_preproc.append(nn.ReflectionPad2d(2))
|
|
layers_preproc.append(smoothing.AverageSmoothing2D(channels=3+c_dim, kernel_size=21))
|
|
self.preprocessing = nn.Sequential(*layers_preproc)
|
|
|
|
def forward(self, x, c):
|
|
# Replicate spatially and concatenate domain information.
|
|
# Note that this type of label conditioning does not work at all if we use reflection padding in Conv2d.
|
|
# This is because instance normalization ignores the shifting (or bias) effect.
|
|
c = c.view(c.size(0), c.size(1), 1, 1)
|
|
c = c.repeat(1, 1, x.size(2), x.size(3))
|
|
x = torch.cat([x, c], dim=1)
|
|
|
|
# print(x.shape)
|
|
x = self.preprocessing(x)
|
|
# print(x.shape)
|
|
return self.main(x), x[:,:3]
|
|
|
|
|
|
def avg_smoothing_filter(channels, kernel_size):
|
|
kernel = torch.ones((channels, 1, kernel_size, kernel_size)) / (kernel_size * kernel_size)
|
|
return kernel |